ABSTRACT. Different methods for constructing a public transport (PT) network that optimally serves trips defined by the OD matrix are proposed, and, typically, the generated network serves the demand better than the real-world network (Guihaire & Hao, 2008; Hoa et al, 2021). Yet the use of these methods in practice is very limited, if ever. Why do stakeholders ignore this huge bulk of knowledge? On the methodological side, this may be attributed to the unavoidable disregard of the real-world effects, like the comfort of the bus trip itself, or users’ readiness to change their travel habits. On the practical side, however, the reason for the stakeholders' skepticism may be the inherent sluggishness of the PT system: The price of a mistake is very high, any change must be accepted at many levels, each new line demands an essential budget, the time to assess line(s)’ effectiveness is months, and if lines become unnecessary, their cancellation may be politically or socially impossible. All these phenomena prevent the PT system from following the evolving demand. Our approach to the PT development adopts these inherent limitations on the PT network development: instead of delivering a “silver bullet" solution, we propose a gradual co-adaptation of the PT network development following land-use development, inter-urban migration, change of travelers’ demand, and modal preferences. Below we present the major steps of this approach for the transportation system with private cars, buses, minivan services, and shared taxis. Users’ adaptation to the PT network changes is modeled with the MATSim – Multi-Agent Transportation Simulation framework (Horni et al, 2016). MATSim generates daily multi-modal transportation dynamics based on the data on origin-destination flows, car ownership, users’ non-travel activities, and individual (dis)utility of every transportation mode. In the simulation, MATSim agents (travelers) selfishly choose the mode, start time, and route of each trip between activities trying to maximize their overall daily utility. This daily process is repeated, and, in each iteration, 5-10% of travelers try to modify their plans to match the emerging traffic conditions. The better plans are adopted, and the worst are forgotten until the system reaches user equilibrium (UE). The PT network responds to the evolving demand iteratively, in stepwise iterations: 1. Identify PT users whose travel (dis)utility can be lessened in case they use the fastest mode and the shortest path between their origin and destination. 2. Recognize clusters of their possible trajectories and rank these clusters by their potential to be turned into a PT line. 3. Employ an Evolutionary Algorithm (EA) to establish a bus line that serves one or more top clusters and add this line to the network. 4. Simulate adaptation of travelers' population to the new conditions with MATSim. 5. In a new UE, identify lines to cancel and with MATSim, simulate adaptation to the new conditions and the final UE for the next iteration. Steps 1-5 are repeated until the trajectories’ clusters cannot be recognized anymore (Figure 1).

FIGURE 1 HERE

Fig. 1. Simulation of the travel demand and transit network co-adaptation

To elaborate on each step: 1. Potential passengers of the new bus line are sought among the users of shared taxis users and buses, whose travel (dis)utility can be lower compared to the utility of the car trip due to waiting time, transfers, cost, detours, etc. We equip each of these travelers with a virtual car and estimate the set of top possible trajectories (in terms of travel utility) given the current traffic conditions. Next, we search for the spatio-temporal clusters of these trajectories. 2. Clusters are ordered by their importance which is defined by the number of trajectories in each cluster, the length of the cluster in space, and its duration in time. 3. A new bus route may serve either a single cluster, persistent in time and well-shaped in space, or several shorter ones. We employ cross-entropy (CE) EA to find a route candidate. The CE iteration consists of two phases: 1) according to a parametric stochastic mechanism R(p), generate a sample of feasible route candidates; 2) update the parameters p of R(p) based on the generated set of candidates to increase the probability that R(p) will produce “better” candidates. In our case, the candidate represents the route’s operating hours, frequency, and sequence of stops. At each CE iteration, the algorithm selects the most beneficial route alternatives and applies an importance sampling to update the generation mechanism R(p). 4. The CE-constructed route is added to the transit network, and MATSim simulation is employed to reach the UE for it. 5. Finally, the algorithm of line(s) cancellation is employed: A line can be cancelled if almost all its users can be served, without the loss of the level of service, by the rest of the networks’ lines. It is important to note that the planner can intervene at each step of the approach and change, e.g., the clusters to be served by the new line, the line’s shape and timetable, or the lines to cancel. At the conference, we will present the results of the PT network adaptation experiments performed with synthetic and real-world transportation systems.

Guihaire, V., Hao, J-K, 2008, Transit network design and scheduling: A global review, Transportation Research Part A: Policy and Practice, 42(10), 1251-1273 Hoa T.M. Nguyen, Andy H.F. Chow, Cheng-shuo Ying, 2021, Pareto routing and scheduling of dynamic urban rail transit services with multi-objective cross entropy method, Transportation Research Part E: Logistics and Transportation Review, 156, 102544 Horni, A., Nagel, K., Axhausen, K.W. (Eds.), 2016. The Multi-Agent Transport Simulation MATSim. Ubiquity Press

ABSTRACT. Bus scheduling is an important task in public transit planning that entails assigning particular vehicles to planned bus trips. Bus scheduling decisions are made after the line planning and timetabling are made. Bus scheduling is typically repeated daily or weekly based on the availability and maintenance routines of the fleet's vehicles. When this common practice is applied, the bus scheduling solution can not be responsive to information that may be gained in real time about the demand for the service. Hence, regardless of the actual occupancy, the same vehicle type (with the same capacity) is used.

This study focuses on a dynamic version of the bus scheduling problem (DyBSP). In the DyBSP, vehicles are assigned to trips a short time before their beginning based on information on the number of passengers on each bus trip obtained through journey reservations made by the passengers. The fleet is assumed to consist of various vehicle models with different capacities and operational costs. The planner aims to dynamically assign the vehicle so as to minimize the expected operational cost while satisfying the demand of all passengers.

Through numerical experiments based on realistic demand data, we demonstrate that near-optimal dynamic assignment of buses to trips can be particularly beneficial in rural areas when demand is sparse and the daily number of trips in each line is small. Evidence shows that in this market segment, many buses travel with very few passengers and sometimes even empty. From a pure business perspective, such "ghost lines" should be discontinued. However, public transit is considered by many as an essential public service that should be provided to any citizen, including those who live in sparsely populated areas. In these markets, public transit services are highly subsidized. Hence, the public interest is in economizing the service by assigning vehicles of the right size to each trip and canceling trips when no passengers arrive. Reducing the vehicle size may also reduce the negative externalities of buses, such as air pollution, noise, and road wear and tear.

In parallel with the research reported in this talk, we conducted a survey-based usability study to evaluate the respective users' potential acceptance of the proposed journey reservation system and operational mode. This study concludes that such a system, based on a mobile app and supported by a call center service, is now a viable option for bus riders in Israel.

ABSTRACT. 1. Introduction

Shared mobility services represent an innovative transportation paradigm encompassing various modes like carsharing, bike-sharing, ridesharing, and ride-hailing. These services offer manifold environmental, social, and transportation advantages. Notably, they cater to users' mobility needs akin to private vehicles or taxis but at a more economical rate. Ideally, these services seamlessly integrate with public transportation networks to bridge service gaps, extending access to those just beyond the reach of conventional services, whether spatially or temporally. The sharing is performed by enabling multiple users to make use of the same resource, either sequentially (vehicle-sharing) or simultaneously (ride-sharing). This work focuses on ride-pooling services, a form of ridesharing wherein either the city or a private enterprise manages a fleet of vehicles (often shuttles) to provide on-demand transit. Typically, a ride-pooling service delineates a service area within which users can request trips to and from various locations. Operators must decide which requests to accept and which vehicles to assign them to. A sustainable service will pool a substantial portion of rides without significantly extending journey times due to detours and intermediate stops. Around 400 ridesharing services operated globally at the end of 2020 (Foljanty, 2020). However, many services have ceased operations as operators struggled to maintain both profitability and service quality. A primary challenge in ride-pooling service design lies in harmonizing various strategic and tactical planning decisions concurrently. Crucially, the overarching challenge is to entice private vehicle users to opt for ride-pooling without shifting those already utilizing sustainable modes like walking, cycling, and public transit. The work presented herein is part of the Seamless Shared Urban Mobility project (www.sum-project.eu), funded by Horizon Europe. The project's objective is to transform existing transportation networks by integrating new shared mobility solutions with public transit. It brings together 30 partners from academia, city government, public transit operators, and the private sector and establishes nine living labs, one in Jerusalem and eight in European cities. Each living lab focuses on a specific shared mobility challenge. The Jerusalem Living Lab team consists of AUTOlab at Tel Aviv University, Jerusalem Municipality, the Jerusalem Transportation Master Plan Team, and the data consultancy, Sigma6. The partnership will reevaluate the potential of ride-pooling services in Jerusalem following the closure of "TikTak", which operated fifty vehicles from 2020 to 2022 but proved inefficient. The team will explore operational scenarios that leverage existing resources to enhance and extend the public transit network, considering the service area, local commute pattern, and places of interest across several neighborhoods. Focusing on the Yuvalim-Ganim area, our primary objective is to answer this question: given the limited resources available for an on-demand transit service (including the number of vehicles and driver hours), how should the service be designed and operated to maximize its effectiveness?

2. Methodology

We are developing a holistic framework for managing ride-pooling services, comprising three key elements: mode choice, simulation, and optimization. While many studies in mode choice rely on external data, such as mobility preference surveys, our framework takes a novel approach. It integrates data from simulated ride-pooling operations into the mode choice model, leveraging a high-resolution simulator called FleetPy (Engelhardt et al., 2022). These two components form a feedback loop: the mode choice model predicts how many travelers will select the ride-pooling service, FleetPy simulates the service and generates service quality metrics, these metrics feed back into the mode choice model, changing the modal split for the next iteration. When this feedback loop converges, it offers an evaluation of a given service design. The outer layer of the framework concerns the optimization of the service design. Applying a simulation-based optimization approach, this layer generates each service design, calls the simulation loop, and receives an evaluation. A service design consists of defining: the service area, the allowed pickup and drop-off locations, the fleet composition (size, vehicle types), pricing schemes, and operational guidelines such as maximum allowable detour time or distance. All of these definitions significantly influence the pooling potential and trip acceptance, which we seek to optimize. The goal is to explore the immense parameter space intelligently to uncover promising settings. The challenge lies in balancing the computational cost of each evaluation against the desire to find good solutions with high confidence. Approaches to overcoming this challenge include parallel processing, switching between approximate models and full simulations, and varying the number of repetitions supporting each evaluation. Even with these approaches, we will only be able to evaluate a small number of possible designs, necessitating the development of efficient search heuristics. While our current efforts are centered around Jerusalem, the methodology we employ is applicable beyond this cityscape. The Jerusalem Living Lab team has recently completed the initial phase of data gathering, employing a multifaceted approach that incorporates mobility surveys, historical transit data, and the utilization of the city’s GIS systems. Simultaneously, our partner cities have been engaged in similar data collection efforts. As we progress further into the experiment, our intention is to validate our framework by incorporating data from other prominent urban centers involved in the project, including Munich and Krakow.   References Engelhardt, R., Dandl, F., Syed, A.-A., Zhang, Y., Fehn, F., Wolf, F., Bogenberger, K., 2022. FleetPy: A Modular Open-Source Simulation Tool for Mobility On-Demand Services. http://arxiv.org/pdf/2207.14246v1. Foljanty, L., 2020. Mapping the global on-demand ridepooling market. https://lukas-foljanty.medium.com/mapping-the-global-on-demand-ridepooling-market-f8318de1c030. Accessed 10 May 2023.

ABSTRACT. Robots operate in the real world, which is hybrid, i.e. comprised of continuous and discrete properties, uncertain, constrained, non-linear, and often cooperation or at least synchronization with other agents, human and robotic is required. A long time horizon is usually required for an agent to reach its goal, accumulating errors and making discretization impossible. Each of these separately poses a major challenge for autonomous behavior. Robots must be able to come up with long-term plans in the face of these challenges in order to reach autonomy. These problems have been addressed by a variety of research communities, such as in control, automated planning, machine learning, and robotics, each with its own strengths and weaknesses.

In this work, we address a mixed discrete-continuous hybrid planning and control problem for multi-agent in a centralized setting. This setting also exhibits temporal constraints such as deadlines and time windows on the agents' activities, and spatial constraints such as proximity and regional on the agents themselves. There are two parts to the problem: a discrete part represents the abstract state of the world and its happenings, and a continuous part represents the dynamics of the agents operating in the world. The dynamics of the agents is modeled by nonlinear differential equations of second order, with input and state constraints.

We propose to address this problem with an iterative approach. A relaxed problem is solved using an off-the-shelf hybrid planner capable of handling first-order linear models, followed by verification using a non-linear second-order model using tools from optimal control theory. If the solution is verified then a trajectory optimization state is invoked for the full problem, and the plan, schedule, and trajectory are returned as the complete solution to the problem. If the verification fails, as the relaxed solution is infeasible in the full setting, a new solution based on the same structure of the relaxed solution is searched via continuous non-linear optimization using deep neural networks. When there is no hope of finding a relaxed solution, using logic tools, the problem is transformed into a new problem, eliminating the relaxed solution from the solution space, and a new iteration begins. As an extension of this, a setting with disturbances was also taken into consideration, and verification and trajectory optimization stages were developed.

Since we do not implement any discretization, our method can work over long continuous horizons. Concurrency and synchronization are allowed, as the solutions are sound for a real system due to the realistic non-linear dynamic models. Moreover, our iterative approach can efficiently balance between finding good candidate plans with relaxed planners for efficiency, and validation with non-linear realistic models for the complex verification only.

(Funded by ISTRC Proposal: Hybrid Planning for Last Mile Logistics Using Trucks and Drones)

ABSTRACT. As the volume of e-commerce is growing rapidly, rising traffic levels, pollution, and increasing customer demands are making the delivery process increasingly difficult. Customers expect their delivery to arrive faster and within increasingly lower time windows, and last-mile logistics are becoming an important and costly part of the delivery process. Fully autonomous last-mile delivery robots are a promising solution to provide a cost-effective sustainable solution for the delivery to reach its destination. However, while these robots have fully autonomous capabilities, the growing consensus is that at least for the foreseeable future, autonomous robots in an urban environment will routinely encounter situations that they cannot address. A blocked road, bad weather conditions, a congested cross section or interaction with a human may all require the intervention of a remote human operator. The current project aims to examine the requirements and design of a teleoperation interface for a remote human operator in order to be able to support a large number of delivery robots simultaneously. We take a human-machine interface (HMI) perspective by first investigating the requirements, followed by designing, developing, and evaluating user interface methods, and providing guidelines to best support the control of a large number of autonomous remote last-mile delivery robots. As a first step, we aim to understand the various challenges and requirements of last-mile delivery robots, focusing on the teleoperation task. The urban environment is complex, and many unexpected events might require remote human assistance. In order to design a teleoperation interface for an autonomous urban robot, first, a thorough examination of the task is needed. To gather these requirements, we have conducted 15 interviews with experts from both academia and industry who deal with either last-mile urban robotics, teleoperation of robots, or both. The interviews were semi-structured, focusing on the challenges of urban robotic delivery, an elicitation of use cases in which teleoperation is needed, and requirements for the user interface. The interviews were analyzed using a thematic analysis approach. Interviews were first fully transcribed, and using an inductive approach the challenges, tasks, and requirements were gathered based on thematic similarities. Participants emphasized the interaction between the robots and pedestrians as well as interactions initiated by the teleoperator. Other issues that emerged from the interviews included connectivity problems, blocked or disturbed roads, and the teleoperation UI design features. The outcome of this phase consists of a list of possible use cases and challenges that a remote operator might face, as well as concrete guidelines for the design of a user interface for the remote operation of a last-mile delivery robot.

ABSTRACT. The promise of fully autonomous vehicles has faced repeated delays, with widespread availability now projected to be limited to specific routes or geographic areas in the near future. Instead, conditional or semi-autonomous systems (SAE Level 3) are being introduced, where the vehicle can drive autonomously under limited conditions but still requires a human driver to take control when needed.

This paper examines the regulatory and liability challenges posed by these semi-autonomous vehicles. Existing laws have gaps in addressing the ambiguous responsibility of the human driver in the shared control environment of Level 3 autonomy. Factors like fatigue, lack of situational information, and over-reliance on the autonomous system diminish a human driver's ability to intervene effectively.

Current negligence and product liability tort doctrines struggle to allocate liability in semi-autonomous scenarios due to issues of foreseeability, the unpredictability of AI decision-making and shared human-machine control. No-fault insurance schemes also face hurdles in determining coverage and premiums for semi-autonomous risks.

To address these gaps, the paper proposes defining a new legal "supervisor" role for human drivers of semi-autonomous vehicles, with liability dependent on providing adequate alerts and road/system status data from the autonomous driving system. Delineating human and machine responsibilities through regulation can help align legal frameworks with the realities of human-AI interaction while promoting the safety benefits and continued development of vehicle autonomy technology.

ABSTRACT. The panel will address the key findings of the research team with scholars from the fields of law and ethics (and from the Univ of Haifa, the Technion, practice and others) will provides results, findings and insights from their studies. Sub-topics - TBD.

ABSTRACT. The term concrete mix design may be defined as the process of selecting suitable ingredients for construction of concrete and to determine their relative amounts with the objective to produce a concrete mix having the required, strength, workability, and durability in as economical manner as possible. The plastic state and the hardened state are the two states governing the proportioning of ingredient of the concrete mix. If the concrete (plastic state) is not workable, it becomes difficult to be properly placed and compacted. Therefore, the property of workability becomes of vital importance. The present work is based on the effect of addition of rubber crumbs on Portland cement concrete. The present experiments are based on studying the effects of replacement of 4%, 8%, and 12% of cement by rubber crumbs on Portland cement concrete. Number of cubes were produced and the densities and compressive strengths were evaluated at 7, 14, and 28 days.

Basalt fiber-reinforced concrete (BFRC) is relatively a new type of fiber-reinforced concrete, which has demonstrated good mechanical performance. Several types of fibers such as steel, glass, and carbon fibers, are used to improve the performance of plain concrete. However, basalt fibers (BF) are being considered superior to these because of their comparable mechanical strength, higher durability than glass fibers, lower cost than carbon fibers, sustainability due to abundant raw material, and environment-friendly production process. Hence, the number of studies on BFRC are increasing over the years. This experiment covers the properties of BF and BFRC. The effects of BF length and dosage are discussed in detail in term of fresh, hardened, and durability properties of BFRC, followed by highlighting some areas for future research. It is concluded that BF can potentially replace the other conventional fibers which are being used in the industry. Accordingly, the standard notched Compressive test have been carried out to assess the basalt fibre-reinforced concrete's residual compressive strength with an additional 0.125%, 0.1%, 0.2%, and 0.3% of volume fraction of basalt.

ABSTRACT. Anion-exchange membrane fuel cells (AEMFCs) have garnered significant interest within the research community driven by the prospect of utilizing platinum-group-metal (PGM)-free catalysts, positioning them as a highly promising alternative to conventional proton-exchange membrane fuel cells [1]. However, achieving complete PGM-free AEMFC is challenging in terms of performance and durability. Among the various investigations presenting performance data for AEMFCs, fewer than 10% rely on PGM-free catalysts, and among those, only a few studies are entirely devoid of PGM-based catalysts [2]. Among the very few hydrogen oxidation reaction (HOR) catalysts that have been developed, Ni-based catalysts have shown remarkable ex-situ activity for the HOR. However, their in-situ activity in AEMFC is significantly lower due to instability in alkaline medium. This study addresses this gap by showcasing the performance of completely PGM-free AEMFCs with Ni and Co based anode and cathode catalysts. A unique activation method was developed to increase the HOR activity of the Ni-based anode catalyst [3]. By employing this activation method, we observed significant improvement in the catalytic activity of these catalysts within the AEMFC environment. The overall results indicate an enhancement in cell performance compared to non-activated catalysts. Our findings demonstrate a substantial improvement in cell performance as the temperature rises from 60°C to 116°C, achieving a worldwide record-high performance [3]. Another challenge facing this technology is the issue of carbonation, which limits its performance to pure oxygen rather than ambient air as the cathode feed gas. We evaluated the performance of the developed PGM -free AEMFC under different cathode gas feed and showed good stability of the AEMFC in ambient air at high temperature. These improvements are attributed to the enhanced electrochemical kinetics of the PGM-free catalysts, improved mass transport across the cell as well as diminished effect of CO2 at high temperature.

ABSTRACT. As the attention dramatically grows on reducing the overall carbon footprint of transportation, novel energy technologies have flourished in recent years to achieve zero-carbon emissions. In the undergoing green revolution, alkaline electrochemical energy devices such as anion-exchange membrane (AEM) fuel cells[1–3] and AEM water electrolyzers[4,5] have emerged as the next-generation solutions to address environmental issues and energy scarcity. AEMs, as their key components, play an imperative role as the medium to transport the hydroxide ions (OH‾) from cathode to anode[6,7]. In conventional membrane synthesis, most efforts have been dedicated to pursuing optimal ion exchange capacity (IEC) and hydroxide conductivity. However, these inevitably lead to unfavorable swelling behavior and reduced mechanical stability[8]. While fuel crossover behavior is enlarged in a swollen membrane, insufficient strength easily leads to defects such as cracks, tears and pinholes in the membranes, subsequently causing degradation of device performance and ultimately leading to device failure[9]. Therefore, achieving the trade-off between durability and electrochemical properties is a foremost task for the widespread application of AEMs. Among various design strategies to realize high mechanical stability, pore-filling technology is notably considered as a highly targeted and effective solution[10]. AEM developed by a pore-filling method is incorporated with a porous membrane which polyelectrolyte is anchored within. While the impregnated polyelectrolyte contributes to the high conductivity, the porous substrates like Polyethylene (PE) serve as robust frameworks to endow excellent mechanical strength. In this study, polystyrene-block-poly(vinyl benzyl chloride) (PS-b-PVBC) was employed in pore-filling AEMs. PE with different thicknesses were applied as substrates to investigate the effect. To fabricate a homogenous pore-filling AEM (PFAEM), a monomer solution was prepared first by mixing styrene and vinyl benzyl chloride as monomers, divinylbenzene as a crosslinker, and benzoyl peroxide as an initiator. Subsequently, the prepared solution was impregnated into a PE substrate with vacuum assistance, followed by heating in the oven at 60℃ for 18h for polymerization. To conduct amination, the pore-filling membrane was immersed in trimethylamine solution at room temperature for 48h. After washing with water, the PFAEM was kept in distilled water for further characterization. Physical-chemical properties including IEC, OH‾ conductivity, water uptake, swelling ratio and mechanical strength were evaluated. Pore-filled AEM showed a moderate IEC of 1.2 mmol/gr. The OH‾ conductivity is 31.2 mS/cm, which was tested in water at 80℃. While the PFAEM is as thin as 17um, it achieves superior mechanical strength over 160 MPa and significantly suppresses the area swelling behavior. In the next step, further modifications to the fabrication process are necessary to obtain a higher filling ratio and better electrochemical properties. In addition, the comparison is expected to be discussed with the casting membranes prepared by the same polyelectrolytes. The performance of pore-filling AEMs in fuel cells and water electrolyzers needs to be tested.

Reference [1] D.R. Dekel, I.G. Rasin, S. Brandon, Predicting performance stability of anion exchange membrane fuel cells, J. Power Sources. 420 (2019) 118–123. https://doi.org/10.1016/J.JPOWSOUR.2019.02.069. [2] S. Gottesfeld, D.R. Dekel, M. Page, C. Bae, Y. Yan, P. Zelenay, Y.S. Kim, Anion exchange membrane fuel cells: Current status and remaining challenges, J. Power Sources. 375 (2018) 170–184. https://doi.org/10.1016/J.JPOWSOUR.2017.08.010. [3] D.R. Dekel, Review of cell performance in anion exchange membrane fuel cells, J. Power Sources. 375 (2018) 158–169. https://doi.org/10.1016/j.jpowsour.2017.07.117. [4] M. Chatenet, B.G. Pollet, D.R. Dekel, F. Dionigi, J. Deseure, P. Millet, R.D. Braatz, M.Z. Bazant, M. Eikerling, I. Staffell, P. Balcombe, Y. Shao-Horn, H. Schäfer, Water electrolysis: from textbook knowledge to the latest scientific strategies and industrial developments, Chem. Soc. Rev. 51 (2022) 4583–4762. https://doi.org/10.1039/d0cs01079k. [5] C. Santoro, A. Lavacchi, P. Mustarelli, V. Di Noto, L. Elbaz, D.R. Dekel, F. Jaouen, What is Next in Anion‐Exchange Membrane Water Electrolyzers? Bottlenecks, Benefits, and Future, ChemSusChem. 202200027 (2022). https://doi.org/10.1002/cssc.202200027. [6] D.R. Dekel, S. Willdorf, U. Ash, M. Amar, S. Pusara, S. Dhara, S. Srebnik, C.E. Diesendruck, The critical relation between chemical stability of cations and water in anion exchange membrane fuel cells environment, J. Power Sources. 375 (2018) 351–360. https://doi.org/10.1016/j.jpowsour.2017.08.026. [7] Y. Zheng, U. Ash, R.P. Pandey, A.G. Ozioko, J. Ponce-González, M. Handl, T. Weissbach, J.R. Varcoe, S. Holdcroft, M.W. Liberatore, R. Hiesgen, D.R. Dekel, Water Uptake Study of Anion Exchange Membranes, Macromolecules. 51 (2018) 3264–3278. https://doi.org/10.1021/ACS.MACROMOL.8B00034/ASSET/IMAGES/LARGE/MA-2018-00034C_0014.JPEG. [8] J.-H.; ; Choi, P.K.T.; Nguyen, D.-J.; Kim, Y.S. Yoon, G. Das, J.-H. Choi, P. Khanh, T. Nguyen, D.-J. Kim, Y.S. Yoon, Anion Exchange Membranes for Fuel Cell Application: A Review, Polym. 2022, Vol. 14, Page 1197. 14 (2022) 1197. https://doi.org/10.3390/POLYM14061197. [9] M.P. Rodgers, L.J. Bonville, H.R. Kunz, D.K. Slattery, J.M. Fenton, Fuel cell perfluorinated sulfonic acid membrane degradation correlating accelerated stress testing and lifetime, Chem. Rev. 112 (2012) 6075–6103. https://doi.org/10.1021/CR200424D/ASSET/IMAGES/MEDIUM/CR-2011-00424D_0020.GIF. [10] T. Yamaguchi, F. Miyata, S.I. Nakao, Pore-filling type polymer electrolyte membranes for a direct methanol fuel cell, J. Memb. Sci. 214 (2003) 283–292. https://doi.org/10.1016/S0376-7388(02)00579-3.

ABSTRACT. The atrocious attacks on October 7th, 2023 will be forever remembered as one of darkest days in the modern state of Israel’s history. In the transportation sector, it is a stark wakeup call to safeguard against future disruptions and disasters. Today, the transportation sector in Israel is heavily reliant on fossil fuel products (>70%) which are largely sourced from the United Arab Emirates (UAE) and Iraq and from Egypt. Given the volatile political situation, it is imperative that the state of Israel diversifies it’s energy portfolio with alternative energy solutions to protect and grow the economy. Hydrogen is a clean fuel with a high energy density that can be produced from a variety of domestic resources. It can be used as a fuel in the transportation sector via fuel cell technology, however, current fuel cell technology is hindered by the cost of the expensive components to facilitate the acidic operating environment. Anion exchange membrane fuel cells (AEMFCs) are a newer technology complemented by a more favorable alkaline operating environment. The technology is very promising in terms of performance, durability and cost compared to the incumbent, but still has some challenges to overcome. For instance, while substantial progress has been made in terms of the performance of the technology over the last few years, the rapid translation of electrocatalysts from fundamental studies to high-performance devices could further facilitate their development, integration and commercialization into state-of-the-art devices. In our collaborative work, we investigate devices that utilize ionomerless ultra-low-loading electrocatalysts synthesized by co-physical vapor deposition. Our systematic AEMFC experiments demonstrate comparable activity trends to those in three-electrode cells and ultimately satisfy the US Department of Energy’s platinum group metal (PGM) loading and cost targets. With these advancements, we visualize the reality of bringing Israel one step closer to energy independence and security. Once you have energy, you have everything!

ABSTRACT. Anion-exchange membrane fuel cell (AEMFC) is one of the most affordable and efficient technologies that can potentially revolutionize energy storage and delivery. However, during operation, AEMFCs are hampered by the requirement of proper water management, since the oxygen reduction reaction (ORR) consumes water on the cathode side.[1][2] Herein, we explore the effect of water on the cathode electrocatalyst performance as well as the mechanism of the ORR at different hydration levels (λ). In this study, cyclic voltammetry (CV) curves under Ar and O2 environments are evaluated. CV curves under O2 and different λ levels demonstrated that the cathodic peak voltage of ORR negatively shifts with increase in alkali concentration. This seems to be due to the Nernst shift of the H2O/H2 and O2/OH- equilibrium potentials.[3] Linear sweep voltammetry (LSV) curves under flowing of O2 and different water content in the electrolyte shows that the current density drops when the alkali concentration increases. This is due to the decrease in O2 solubility and diffusion coefficient, and the increase in KOH solution viscosity.[3] The overpotential increases with increasing KOH concentration, which is probably due to the changes in conductivity of the electrolyte.[4]

ABSTRACT. The rapid growth of electric cars in recent years has spurred the demand for swift, efficient, and user-friendly charging solutions. Among the emerging options in this realm is wireless charging, offering users high levels of convenience and opening up new market segments. However, one of the primary challenges facing wireless charging is maintaining high efficiency, especially compared to traditional wired methods.

A wireless charging system comprises two main components: the transmitter (AC-AC converter), which in our system relies on wireless inductive power transfer, and the receiver. Our project primarily focuses on the receiver, which is a rectifier circuit. A basic rectifier consists of a diode bridge (4), implementing a mathematical operation of absolute value on the input waveform.

In the basic rectifier (diode bridge) power losses across the rectifier may become significant, leading to lower overall system efficiency, and increasing the amount of required cooling (alongside system weight and form factor).

To reduce losses, the diodes may be replaced by switches with low conduction resistance - resulting in improved efficiency.

This active rectifier (AR) topology has the added benefit of controlling output power flow by utilizing methodologies such as pulse width modulation and phase shift control. In spite of these advantages, this solution requires additional electronic components, raising system cost and complexity. Furthermore, extra care needs to be invested in system control (in order to correctly synchronize switch operation to input waveforms) and protection circuitry in order to refrain from two transistors of the same half bridge reaching ON sate simultaneously, short circuiting the output capacitor.

A middle ground between full bridge diode rectifier and AR (active rectifier) may be attained by replacing only lower leg diodes with switches, resulting in the semi-active rectifier (SAR) topology. In comparison to a diode rectifier, SAR has lower conduction losses, smaller form factor and retains ability of AR to control output power flow, while slightly increasing system complexity and cost. In comparison to AR, danger of output capacitor short circuit is eliminated, and system cost is reduced while efficiency and form factor performance slightly deteriorate.

The SAR is fed by sinusoidal current source (with voltage imposed by the converter) which is highly common in applications such as wireless power transfer systems operating in load independent constant output current mode. In our circuit operation, a key principle is the concept of ZVS (Zero Voltage Switching), aimed at reducing losses and increasing efficiency. ZVS entails sensing the voltage across the transistor, equating it to zero, and switching it ON at that point. This method effectively reduces losses and increases efficiency, as confirmed by analytical calculations, PSIM simulations, and basic experiments.

This project serves as our final graduation project and continues to evolve over time. We aim to have a physical circuit onboard, including control via microcontroller, by the time of the conference. We are optimistic about the outcomes and look forward to sharing our progress and findings with the research community.

ABSTRACT. Nickel-based catalysts for hydrogen oxidation reaction in alkaline medium

Alexander Baranov a, Dario R. Dekel a,b

a The Wolfson Department of Chemical Engineering, Technion – Israel Institute of Technology, Haifa, Israel b The Nancy & Stephen Grand Technion Energy Program (GTEP), Technion – Israel Institute of Technology, Haifa, Israel E-mail: dario@technion.ac.il Keywords: nickel; electrocatalysts; hydrogen oxidation reaction; alkaline.

Anion-exchange membrane fuel cell (AEMFC) technology offers a highly efficient and environmentally friendly way to generate energy, ideally integrating into the alternative energy concept1. Over the past decade, in the course of intensive fundamental research and development of materials, AEMFCs have achieved new records in both performance and durability. With the ability to use low-cost catalysts2, membranes3, and bipolar plates, AEMFCs are a cost-effective choice for a wide range of applications, such as fuel cell-powered electric vehicles. Despite the above attractive advantages of AEMFCs, current cells still rely on platinum group metal (PGM) catalysts, which is far from practical application and does not fully realize the economic benefits of this technology. Therefore, the transition to the use of PGM-free catalysts is the next goal of AEMFCs. To date, significant progress has been made in the research of PGM-free catalysts for the oxygen reduction reaction at the AEMFC cathode, some of which have demonstrated good catalytic properties comparable to PGM catalysts4. The development of PGM-free catalysts for the anodic hydrogen oxidation reaction (HOR) has lagged behind other AEMFC components and represents a major bottleneck for the technology5. This research aims to fill this gap by developing new nickel-based catalysts. The catalysts were synthesized through annealing of metal-organic frameworks (MOFs). These MOFs, based on nickel and iron and linked by benzene-1,3,5-tricarboxylic acid or 2-aminobenzene-1,4-dicarboxylic acid, were prepared using the solvothermal method in dimethylformamide. Subsequent annealing of the compounds in a mixed atmosphere of hydrogen and nitrogen led to the production of catalysts with metal nanoparticles distributed on a carbon support. The activity of these catalysts towards alkaline HOR was studied using a three-electrode cell equipped with a rotating glass carbon electrode. The measurements were carried out using cyclic voltammetry (CV), linear sweep voltammetry (LSV), and electrochemical impedance spectroscopy (EIS). The optimization yielded a catalyst with a high nickel mass fraction of 80% and simultaneously, a high electrochemical surface area of 54.2 m2/gNi. This corresponds to nickel nanoparticles with an average size of 12.4 nm. The HOR current density reaches 1.7 mA/cm2 at 50 mV vs RHE and 2500 RPM, which is 10 times higher than that of the commercial 80% Ni/C Premetek catalyst used for comparison (0.15 mA/cm2). The catalyst exhibits a specific exchange current density of the HOR i0 = 29.6 μA/cm2ECSA, which indicates good surface activity.

References 1. Dekel, Dario R. “Review of Cell Performance in Anion Exchange Membrane Fuel Cells.” Journal of Power Sources 375 (January 2018): 158–69. https://doi.org/10.1016/j.jpowsour.2017.07.117. 2. Davydova, Elena S., Sanjeev Mukerjee, Frédéric Jaouen, and Dario R. Dekel. “Electrocatalysts for Hydrogen Oxidation Reaction in Alkaline Electrolytes.” ACS Catalysis 8, no. 7 (July 6, 2018): 6665–90. https://doi.org/10.1021/acscatal.8b00689. 3. Gottesfeld, Shimshon, Dario R. Dekel, Miles Page, Chulsung Bae, Yushan Yan, Piotr Zelenay, and Yu Seung Kim. “Anion Exchange Membrane Fuel Cells: Current Status and Remaining Challenges.” Journal of Power Sources 375 (January 31, 2018): 170–84. https://doi.org/10.1016/j.jpowsour.2017.08.010. 4. Singh, Ramesh K., John C. Douglin, Lanjie Jiang, Karam Yassin, Simon Brandon, and Dario R. Dekel. “CoOx-Fe3O4/N-rGO Oxygen Reduction Catalyst for Anion-Exchange Membrane Fuel Cells.” Energies 16, no. 8 (April 13, 2023): 3425. https://doi.org/10.3390/en16083425. 5. Dekel, Dario R. “Review of Cell Performance in Anion Exchange Membrane Fuel Cells.” Journal of Power Sources 375 (January 2018): 158–69. https://doi.org/10.1016/j.jpowsour.2017.07.117.

ABSTRACT. Carbon dioxide (CO2) emissions from transportation are a significant contributor to global climate change, primarily due to the combustion of fossil fuels in vehicles [1]. In response to the urgent need to mitigate these emissions, there's a growing focus on implementing technologies for carbon capture and storage (CCS) within the transportation sector. These technologies, including advancements like direct air capture, aim to capture CO2 emissions from vehicle exhaust before they are released into the atmosphere [2]. By integrating CCS systems into transportation infrastructure, we can effectively reduce the environmental impact of vehicle emissions and curb the harmful effects of CO2 on the climate. Additionally, advancements in automotive technology, particularly in Fuel Cell Electric Vehicles (FCEVs), emphasize the importance of separating CO2 from the air stream before it enters the fuel cell. This process not only enhances the efficiency of FCEVs but also contributes to overall emissions reduction efforts in the transportation industry. Through innovative solutions like CCS and advancements in vehicle technology, we can work towards a more sustainable and environmentally friendly transportation system [3]. Electrochemical separation can be employed for the purification of CO2 from gas mixtures, using oxygen reduction reaction )ORR( and hydrogen oxidation reaction )HOR( as the driving force for CO2 hydration hydration and subsequent separation using only a small input of electricity [4]. However, this technology still utilizes very rare and expensive platinum group metals (PGM) as catalysts at the cathode, rendering it economically unfeasible for large-scale use. Overall, none of the currently available technologies for the separation of CO2 provide continuous, dependable, and cost-effective purification of carbon dioxide. Meanwhile, there is a growing demand for such technology across various industries. This research aims to investigate the potential of an electrochemical cell using an Anion Exchange Membrane (AEM) for CO2 separation, which we would like to call the Carbon Dioxide Separation Cell (CDSC). The innovation of this cell lies in the use of low-cost catalysts with maximum efficiency on the cathode side . As will be described below, literature studies so far show the use of PGM catalysts on both electrodes to CO2 capture. Therefore, while using PGM-free at the cathode, at the same time we will use a PGM catalyst at the anode side (exp: PtRu/C) because the sluggish kinetics of HOR in the alkaline medium seems very much challenging even for PGM catalysts [5]. Over the past decade, there has been significant progress in AEMs, primarily driven by their application in AEM Fuel Cells (AEMFCs). This progress has entailed the development of new membrane materials and PGM-free catalysts [6], significantly lowering the cost of this technology. AEMs have exhibited excellent hydroxide ion conductivity and relatively good chemical stability, thereby substantially improving the performance and durability of AEMFCs. In principle, by adapting some of the electrodes of an AEMFC [7], and optimizing the operating conditions for a different application, it seems possible to create an electrochemical cell capable of functioning as an electrochemical CO2 separation device.

References [1] Hao Yu, Hao Chen, Yi-Ming Wei, Yi-Mei Li. "The influence of climate change on CO2 (carbon dioxide) emissions: an empirical estimation based on Chinese provincial panel data." Journal of cleaner production 131 (2016): 667-677. doi: https://doi.org/10.1016/j.jclepro.2016.04.117. [2] Alper, Erdogan, and Ozge Yuksel Orhan. " CO2 utilization: Developments in conversion processes."Petroleum“3.1 (2017): 109-126. doi: https://doi.org/10.1016/j.petlm.2016.11.003. [3] Zhu, Qian. "Developments on CO2-utilization technologies."Clean Energy 3.2 (2019): 85-100.‏ doi: https://doi.org/10.1093/ce/zkz008. [4] Wang Xiaoguang, William Conway, Robert Burns, Nichola McCann, Marcel Maeder. "Comprehensive study of the hydration and dehydration reactions of carbon dioxide in aqueous solution." The journal of physical chemistry A 114.4 (2010): 1734-1740.‏ doi: https://doi.org/10.1021/jp909019u. [5] Martinez Ulises, Siddharth Komini Babu, Edward F. Holby, Hoon T. Chung, Xi Yin, Piotr Zelenay. "Progress in the development of Fe‐based PGM‐free electrocatalysts for the oxygen reduction reaction." Advanced materials 31.31 (2019): 1806545.‏ doi: https://doi.org/10.1002/adma.201806545. [6] Md. Mosaddek Hossen, Md. Shamim Hasan, Md. Riajul Islam Sardar, Jahid bin Haider, Mottakin, Kaido Tammeveski, Plamen Atanassov, State-of-the-art and developmental trends in platinum group metal-free cathode catalyst for anion exchange membrane fuel cell (AEMFC), Applied Catalysis B: Environmental, Volume 325, 2023, 121733, ISSN 0926-3373, https://doi.org/10.1016/j.apcatb.2022.121733. [7] Matz Stephanie, Brian P. Setzler, Catherine M. Weiss, Lin Shi, Shimshon Gottesfeld, Yushan Yan. "Demonstration of electrochemically-driven CO2 separation using hydroxide exchange membranes." Journal of The Electrochemical Society 168.1 (2021): 014501.‏ doi: 10.1149/1945-7111/abd5fe.

ABSTRACT. Carbon dioxide (CO2) emissions are a significant contributor to global climate change, primarily due to the combustion of fossil fuels in vehicles [1]. In response to the urgent need to mitigate these emissions, there's a growing focus on implementing technologies for carbon capture and storage (CCS) [2]. By applying CCS systems to our infrastructure and environmental control systems, we can effectively reduce the environmental impact of vehicle emissions and curb the harmful effects of CO2 on the climate. Additionally, advancements in automotive technology[3], particularly in Fuel Cell Electric Vehicles (FCEVs), emphasize the importance of separating CO2 from the air stream before it enters the fuel cell. This process not only enhances the efficiency of FCEVs but also contributes to overall emissions reduction efforts in the transportation industry. Through innovative solutions like CCS and advancements in vehicle technology, we can work towards a more sustainable and environmentally friendly transportation system. However, due to the low concentration of CO2 in the ambient air, none of the currently available technologies for the separation of CO2 provide efficient and cost-effective sequestration of carbon dioxide from the ambient air [4].

This research aims to investigate the potential of a novel electrochemical cell using an Anion-Exchange Membrane (AEM) for CO2 separation from ambient air, which we would like to call the Carbon Dioxide Separation Cell (CDSC). This CDSC consists of oxygen reduction reaction )ORR( and hydrogen oxidation reaction )HOR( as the driving forces for CO2 sequestration and subsequent separation using only a small input of electricity and hydrogen consumption. The CDSC is based on our knowledge and know-how on AEMs and AEM fuel cell (AEMFC) technology. In principle, by adapting some of the electrodes of AEMFCs, and optimizing the operating conditions for a different application, it seems possible to create an electrochemical cell capable of functioning as an electrochemical CO2 separation device. In this talk, the novel concept of CDSC will be introduced, and the results of the first tests will be discussed [4-6].

References [1] Hao Yu, Hao Chen, Yi-Ming Wei, Yi-Mei Li. "The influence of climate change on CO2 (carbon dioxide) emissions: an empirical estimation based on Chinese provincial panel data." Journal of cleaner production 131 (2016): 667-677. doi: https://doi.org/10.1016/j.jclepro.2016.04.117. [2] Alper, Erdogan, and Ozge Yuksel Orhan. " CO2 utilization: Developments in conversion processes."Petroleum“3.1 (2017): 109-126. doi: https://doi.org/10.1016/j.petlm.2016.11.003. [3] Zhu, Qian. "Developments on CO2-utilization technologies."Clean Energy 3.2 (2019): 85-100.‏ doi: https://doi.org/10.1093/ce/zkz008. [4] Dekel, Dario R. “Review of Cell Performance in Anion Exchange Membrane Fuel Cells.” Journal of Power Sources 375 (January 2018): 158–69. https://doi.org/10.1016/j.jpowsour.2017.07.117. [5] Singh, Ramesh K., John C. Douglin, Lanjie Jiang, Karam Yassin, Simon Brandon, and Dario R. Dekel. “CoOx-Fe3O4/N-rGO Oxygen Reduction Catalyst for Anion-Exchange Membrane Fuel Cells.” Energies 16, no. 8 (April 13, 2023): 3425. https://doi.org/10.3390/en16083425. [6] Davydova, Elena S., Sanjeev Mukerjee, Frédéric Jaouen, and Dario R. Dekel. “Electrocatalysts for Hydrogen Oxidation Reaction in Alkaline Electrolytes.” ACS Catalysis 8, no. 7 (July 6, 2018): 6665–90. https://doi.org/10.1021/acscatal.8b00689.

ABSTRACT. Amidst rapid urbanisation and increase motorisation, walking remains an important mode of transport in urban areas with far reaching environmental, economic, and social benefits. The frequency with which people walk, the distance for which they may be willing to walk, and the convenience and safety with which they can walk, partly depends on the nature of the road design. In most cases and especially in urban areas of Sub-Saharan Africa (SSA), roads are designed in favour of motorisation even though rising car use is associated with noise and air pollution, negative health impacts, and exclusion of vulnerable populations. This bias is manifested through the road designs that are unattractive in terms of safety, convenience, and beauty for pedestrians as a movement channel and as a place for economic and social activities. Not different from practice, the relationship between road designs and walking has been understudied in urban areas of SSA. This paper addresses this research gap by analysing the impacts of two carefully selected urban roads with distinct designs on residents’ mobility, accessibility, livelihoods, and social interactions in Bamenda, which is part of a larger project involving two other cities (Kigali and Cape Town). The two roads are urban main roads with one having a very basic design without walking facilities except for general street lightening (Mile 2 – Mile 3 Nkwen road) and the other a more complete road design with sidewalks, a constructed high median, and general street lightening (Commercial Avenue Street). Firstly, the two selected roads were assessed using a checklist to identify the road design elements (not) serving pedestrians. Secondly, a survey using an interview guide was conducted with 15 participants living or having shops close to each road, selected using the snowball sampling technique. This is to identify how these roads impact residents’ mobility and livability. Findings show that both roads limit walking mobility and accessibility along and across the streets as well as social interactions between people on both sides despite the existence of sidewalks and designated crossings along the median of the Commercial Avenue Street. This is because the sidewalks on the later street are invaded with business activities leaving little or no space for pedestrians while the high median limits visibility between both sides of the street. However, the Commercial Avenue Street is reported to be relatively better for walking as it has less dust and mud during the dry and rainy seasons respectively while the median act as a buffer and make crossing more organised compared to the Mile 2 – Mile 3 Nkwen Street. While highlighting the impacts of roads on walking and livability, this study calls for more attention towards safe and inclusive road designs for different users and functions and the provision of walking facilities for residents in the city.

ABSTRACT. Micro-mobility riders on urban roads are involved in crashes that may hurt themselves as well as other road users. To improve the safety of micro-mobility riders and the design of better urban infrastructure, it is important to study the behavior and cognitive processes and address the cognition failures of micro-mobility riders. Riding behavior is the result of various factors where in addition to decision-making factors. Safety and hazard perceptions are critical cognitive factors for safe riding. Hence, it is important not only to consider the riding behavior but also to evaluate the safety and hazard perceptions of riders that affect their behavior. While most existing research is focused on observing riding behaviors in the actual environment or through surveys there is less knowledge about the safety and hazard perceptions of riders. To meet the goal of this study, to evaluate the safety and hazard perception of micro-mobility riders, we have developed a simulated testbed and an associated methodology for studying how e-scooter riders explore the urban environment, perceive hazards, and opt for the safest riding positions. Each testbed user is asked to navigate as a virtual rider to the safest riding position in each given urban scenario. We have collected data not only about the final virtual riding positions that reflect the perceived safest riding positions of participants but also about their eye movements and navigation records that reflect their safety and hazard awareness as well as their decision-making processes. The observed decisions in this study indicate that the participants perceive parallel parking as unsafe for riding in the given simulated scenarios.

ABSTRACT. A submission to the student track Background The world population is ageing and will continue to do so in the upcoming decades. Mobility, is a basic human ability, that often declines throughout the ageing process. Mobility is a hallmark of functional ageing, and age-related mobility decline is associated with decreased quality of life and increased morbidity and mortality. Mobility is highly dependent upon the ability to use different transportation modes. So far, age-related mobility decline has been mainly examined from the perspective of driving cessation, and other transport modes remain relatively understudied. Public transport is used by a significant portion of the older adult population as a means of sustaining mobility in later ages, whether among older adults after driving cessation or among those that never drove. Yet, very little is known about age-related decline in the ability to use public transport – public transport use self-regulation and cessation – and its health outcomes. Problem Statement At a certain stage, some older adults might experience a decline in their ability to use public transport, thus adversely affecting their mobility, independence and health. Yet, almost nothing is known about this process. Objective The aim of this study is to describe the process of public transport use self-regulation and cessation among older adults and its impacts on their mobility. Methods 24 semi-structured interviews with older adults living in metropolitan centres in Israel were conducted. Participants were aged between 67-88 and in various stages of public transport use self-regulation and cessation. Thematic analysis was used to identify themes and dimensions related to self-regulation and cessation. Results Public transport use cessation is a non-linear, highly divergent process. Yet, participant widely recognised it, and often regarded it with feelings of fear and anxiety. Interviewees yet to undergo cessation expressed concern about its possible occurrence and impacts on their independence. Interviewees who self-regulate their public transport use or who ceased using public transport altogether report diminishing life-space and a decline in activity participation and quality of life. Fear of falling and past falls on public transport, drivers' attitudes and a declining physical ability are described as the main reasons leading to cessation, with family members and relatives often involved in the decision to cease use. Conclusions Public transport use self-regulation and cessation are important though understudied processes in older adults age-related mobility decline, with adverse impacts on older adults' mobility, independence and quality of life. Further studies are required to assess these processes' epidemiology and to identify possible interventions to allow older adults maintain mobility in later stages of life.

ABSTRACT. New in-vehicle technologies enable road-operators to notify drivers about traffic-related events and recommended behavior. Sinkhole (when a road collapses, and part of it is blocked) and Rubberneck (when queuing traffic develops due to drivers’ curiosity) are two unplanned traffic-related incidents that can quickly escalate into major traffic events. Drivers should be informed on how to behave, as when not appropriately informed, some may panic and react unpredictably, endangering themselves and other road users. The overarching work aims to explore drivers’ preferences for receiving emergency event in-vehicle information, while attempting to fit the information needs to driving traffic-related stress tendency. This exploration utilizes mixed methods of user elicitation techniques towards building a communication model for major traffic events. Specifically, in this presentation I will discuss the outcomes of the on-line questionnaire that investigated the different information needs of drivers in major traffic events, by event type (Rubberneck/Sinkhole), driver stress tendency classification (3 levels of support), and proximity to the event. One hundred eight participants completed an on-line study aimed to test drivers’ message preference (frequency, level of detail, and how to receive it) for both scenarios. In-vehicle notifications were presented in three different frequencies, following each participant’s fit to a specific stress tendency group. The results indicated significant difference in message frequency and level of detail preferences by driver stress tendency and a substantial difference in level of detail between the Rubberneck and Sinkhole scenarios. All respondents were receptive to in-vehicle notification. The findings suggest that road operators should utilize such technologies to address variation in drivers stress tendency and type of emergency event.

ABSTRACT. In the quest for sustainable growth and energy preservation, refining energy expenditure in enclosed areas, especially through advanced ventilation techniques, stands as both a hurdle and a potential. This investigation delves into the nexus of energy efficiency and indoor environment control within domestic garages, illustrating a microcosm of confined spaces. Through an exhaustive case analysis, we crafted and applied an innovative analytical framework aimed at harmonizing ventilation flows with energy consumption, all while adhering to air quality norms. Leveraging empirical data gathering and simulation methodologies, our research pinpoints pivotal methods for boosting ventilation effectiveness, showcases the avenue for considerable energy conservation, and accentuates the significance of custom ventilation solutions in reducing energy footprints. Our discoveries not only add to the dialogue on ecological sustainability but also furnish practical guidelines for the engineering and management of energy- conservative ventilation infrastructures in enclosed environments. This study highlights the essentiality of comprehensive strategies in attaining energy utilization benchmarks and proposes a replicable paradigm for advancing environmental standards in residential and akin sealed spaces.

ABSTRACT. Attention Deficit Hyperactivity Disorder (ADHD) is a developmental neurological disorder characterized by difficulty maintaining focus, multitasking, and intense reactions to situations. Additionally, it can lead to excessive distractibility, fixation on irrelevant details, and impulsive behavior. Over the past decade, awareness of this disorder has increased significantly, especially among adolescents and students, resulting in a growing number of patients diagnosed with ADHD in recent years. Problem definition: As part of the project, we investigated the association between ADHD and involvement in road accidents. We explored and analyzed the road accident data and their severity in relation to attention and concentration disorders. Research Design: We established and distributed an anonymous questionnaire containing all the relevant questions for the study. No personal information about the respondents was collected. The questionnaire focused on demographic data such as age, gender, and place of residence. The respondents were asked about having a diagnosis of ADHD and whether there was any pharmacological treatment and involvement in road accidents. We also gathered data about the severity of the accidents and weather conditions at the time of the accident. 300 participants responded to the survey. Research Questions: Does Attention Deficit Hyperactivity Disorder (ADHD) increase the probability of being involved in road accidents? The primary goal of the study was to determine whether there is a significant correlation between ADHD and accident involvement. We conducted a two-tailed t-test, examining the distribution of road accidents among drivers with and without ADHD. Does taking medication for attention and concentration problems reduce the risk of being involved in road accidents? Individuals diagnosed with ADHD often take medications to manage daily tasks that can be otherwise challenging. We aimed to investigate whether the medications impact driver involvement in road accidents using a two-tailed t-test. Study results: The most common response to the question “Have you been involved in a car accident as a driver?” was "No Accident" (45%), followed by "With ADHD" (16%), "NoLic" (5%), "EasyCrash" (11%), and "HardCrash" (1%). The risk of a minor car accident is 7% for drivers with ADHD without medical treatment, compared to 3% for drivers with ADHD with medical treatment. The risk of a severe car accident is 2% for drivers with ADHD without medical treatment, compared to 1% for drivers with ADHD with medical treatment. Conclusions: The statistical test results and model analysis converged, indicating a positive statistical correlation between Attention Deficit Hyperactivity Disorder (ADHD) and the likelihood of being involved in road accidents. In other words, individuals diagnosed with ADHD have a higher probability of being involved in road accidents. Furthermore, the results demonstrate that pharmacological treatment for attention and concentration disorders decreased the likelihood of involvement in road accidents. Statistically, the cause of the accident did not significantly affect this relationship. Keywords: ADHD, road accidents

ABSTRACT. According to the World Health Organization, approximately 1.2 million people are killed, and up to 30 million are injured in traffic accidents annually worldwide. Several reasons and contributing factors have caused road accidents to be considered a significant public health concern. This study investigated whether the driver's age of 55 and above is an independent risk factor for road accidents. The study primarily focused on various variables that influence the number of road accidents in different age groups. We examined the frequency of accidents by weather conditions, types of accidents, and severity levels by cross-validating each parameter. Driver’s age groups were categorized based on car insurance rates. Problem definition: There is limited research on risks faced by elderly drivers in Israel, particularly as the population ages.

Research Design: The study combined a mixed-methods approach and descriptive and inferential statistical techniques. We created models to estimate our hypotheses and validated them using collected data from the Israel Central Bureau of Statistics for the last ten years. We investigated: • Do weather conditions influence road accident frequency across different age groups? • Which age group is the most dangerous in extreme weather? • Are there significant variations in the types of accidents caused by drivers of different ages? • What is the most common type of accident for elderly drivers?

Study results: Tests and visual representations revealed four critical findings regarding traffic accidents among different age groups in Israel: Seasonal Impact on Accident Frequency: While in Israel, we enjoy mostly sunny weather year-round, the focus on winter months, where precipitation is higher, revealed a significant impact, particularly of the extreme weather, on accident frequency. Age and Accident Severity in Extreme Conditions: We demonstrated that the impact of extreme weather conditions on road accident frequency varies across age groups. Younger age groups demonstrate a higher propensity for severe accidents in extreme weather conditions compared to older age groups (new drivers: 18.8%, young drivers: 17.6%, drivers aged 30: 15.4%, drivers aged 50: 14.2%, drivers aged 70: 16.1% and seniors: 17.4%). Accident Types by Age Group: The study revealed a difference in the types of accidents caused by drivers of different ages. “Injury to pedestrian” and “side to side collision” constitute the most frequent type of accident across all age groups. Compared to other age groups, elderly drivers were more likely to be involved in pedestrian accidents. Most Frequent Accident Type for Seniors: Further analysis showed that "injury to pedestrian" accidents constituted the highest percentage (38.55%) among all other types of accidents for seniors.

Conclusions: The research highlights the importance of considering age-specific risk factors in driving. While considering the worldwide challenge of the aging population, the significance of traffic accidents amongst older adults is rising. Therefore, further research on traffic accidents focusing on older adults may decrease road accident rates and result in more efficient and safe driving.

ABSTRACT. Paratransit transport has become a permanent fixture in many African cities. It is used by millions of people daily, but low and unpredictable service levels, pollution, inadequate safety standards, and road crashes characterize it. Being the only form of motorized transportation for many, paratransit services often provide low-level services and show little respect for their users. Such conditions not only disrespect users but also push them towards even riskier transport alternatives like motorcycle taxis and walking. In many African cities, pedestrian transit emerges as the primary mode of transport, predominantly utilized by the majority, who are low-income individuals. The middle class frequently uses the available paratransit systems whereas regular private vehicular use is predominantly confined to the smallest, affluent segment of the population. The highest road fatalities are among the pedestrians, who often cannot afford to use paratransit, while the paratransit users are also affected by road crashes and other concerns over personal safety and theft. This research uses the concept of dignity to explore increased ridership of paratransit for various groups including, the urban poor, who predominantly walk, and the emerging middle class who aspire to or have acquired car ownership, while addressing the numerous concerns with the existing paratransit systems.Using the concept of dignity, this research explores the experiences of various paratransit actors. Other fields of study have shown that the normative concept of dignity can powerfully highlight unjust relationships and suggest pathways to improvement. Dignity, however, has yet to be employed systematically in the transport literature. Therefore, this study expands upon the application of dignity in various research fields toward a transport adaptation of dignity. The study uses ethnographic methods of observation, ride-a-longs, and post-ride interviews and analyzes rich data on the paratransit experiences of users, drivers, and conductors in Nairobi city. Thematic analysis of the interviews resulted in the classification of experiences as either positive or negative concerning their impact on dignity. Positive themes included acknowledgment and recognition, autonomy, and advocacy, while negative themes included diminishment, labeling, and minimizing. Findings reveal that interactions between actors improve or diminish dignity among actors. The effects of dignity violations indeed reduced ridership among those who could access alternative transportation modes, such as walking and car ownership. Car ownership among the lower-middle class, in some instances, came at the cost of crippling personal loans to avoid the indignities of paratransit. Those who suffered traumatic experiences in paratransit are less likely to use it, and most affected are often the lower income groups who lack alternatives and yet forgo paratransit in favor of walking, which has its precarities. Another alternative sought, is the use of motorcycle taxis, which has the second highest fatalities after pedestrians. Positive experiences, however, increase ridership in vehicles with improved services and improve the relations among actors, their self-perception, and that of actors with whom they interact, providing overall safer and more competitively priced transport options

ABSTRACT. Recent advancements in technology, particularly in computer vision, sensor fusion, and artificial intelligence, have propelled the development of autonomous vehicles (AVs) as a revolutionary mode of transportation. Major automotive manufacturers and emerging startups actively pursue cutting-edge technologies to enable AVs to operate independently. However, despite significant progress, current AVs face challenges in handling various road scenarios autonomously, such as road construction, malfunctioning traffic lights, or busy intersections. For instance, in 2021, industry leaders Waymo and Zoox reported disengagement rates of 7000-8000 miles per disengagement. This underscores the consensus within academic and industrial circles that AVs will encounter traffic situations beyond their autonomous capabilities, necessitating remote human intervention.

Teleoperation emerges as a promising strategy to address these challenges and facilitate the widespread deployment of fully autonomous vehicles on public roads. Teleoperation involves the engagement of a remote human operator (RO) who oversees and guides the vehicle's actions from a distance. When faced with challenging situations, the RO can evaluate and guide the vehicle until the issue is resolved. Several teleoperation systems for AVs are currently operational and undergoing refinement. However, existing solutions primarily rely on a tele-driving paradigm, where the RO directly controls the vehicle using traditional controls such as a steering wheel and pedals. Yet, manual remote control poses significant challenges, including difficulty feeling forces exerted on the vehicle, latency issues due to data transmission delays, high cognitive load of the RO, etc.

Alternatively, the Tele-assistance paradigm suggests that humans can provide high-level guidance to the AV, allowing it to execute low-level maneuvers autonomously. In this approach, the RO communicates high-level directives through a specialized interface rather than directly driving the vehicle. Tele-assistance offers several advantages over manual operation. Firstly, it significantly reduces the duration of teleoperation sessions, as issuing discrete commands is faster than manual steering. Secondly, disconnecting the RO from low-level controls enables the management of heterogeneous vehicles and fleets through universal commands, facilitating smoother adaptation and learning. Thirdly, tele-assistance enhances safety by entrusting low-level maneuvers to the autonomous agent, potentially reducing accidents caused by human errors. Lastly, it may alleviate the cognitive load from ROs compared to tele-driving interfaces, which require high levels of attention.

Our research focuses on the teleoperation of fully autonomous vehicles, aiming to establish a set of high-level commands for tele-assistance and develop a communication framework between AVs and ROs. We conducted simulations of ten scenarios derived from previous studies, eliciting suggestions for high-level commands from seventeen experienced teleoperators. The findings were used to create a scenario-command mapping and a general list of high-level commands categorized into six major categories. Additionally, we designed and evaluated an initial user interface for tele-assistance.

In summary, our work contributes scenario-command mapping, a list of high-level commands, and an initial tele-assistance user interface, laying the groundwork for effective teleoperation of fully autonomous vehicles.

ABSTRACT. Deep reinforcement learning (DRL) has become a dominant deep-learning paradigm for tasks where complex policies are learned within reactive systems. Unfortunately, these policies are known to be susceptible to bugs. Despite significant progress in DNN verification, there has been little work demonstrating the use of modern verification tools on real-world, DRL-controlled systems. In this case study, we attempt to begin bridging this gap, and focus on the important task of mapless robotic navigation — a classic robotics problem, in which a robot, usually controlled by a DRL agent, needs to efficiently and safely navigate through an unknown arena towards a target. We demonstrate how modern verification engines can be used for effective model selection, i.e., selecting the best available policy for the robot in question from a pool of candidate policies. Specifically, we use verification to detect and rule out policies that may demonstrate suboptimal behavior, such as collisions and infinite loops. We also apply verification to identify models with overly conservative behavior, thus allowing users to choose superior policies, which might be better at finding shorter paths to a target. To validate our work, we conducted extensive experiments on an actual robot, and confirmed that the suboptimal policies detected by our method were indeed flawed. We also demonstrate the superiority of our verification-driven approach over state-of-the-art, gradient attacks. Our work is the first to establish the usefulness of DNN verification in identifying and filtering out suboptimal DRL policies in real-world robots, and we believe that the methods presented here are applicable to a wide range of systems that incorporate deep-learning-based agents.

ABSTRACT. In this work, we consider path planning for a team of mobile agents where one agent must reach a given target as soon as possible and the others must accommodate to avoid collisions. We call this practical problem the Single-Agent Corridor Generating (SACG) problem and explore several algorithms for solving it. We propose two baseline algorithms based on existing Multi-Agent Path Finding (MAPF) algorithms and outline their limitations. Then, we present the Corridor Generating Algorithm (CGA), a fast and complete algorithm for solving SACG. CGA performs well compared to the baseline approaches. In addition, we show how CGA can be generalized to address the lifelong version of MAPF, where new goals appear over time.

ABSTRACT. Autonomous driving is a major modern industrial revolution transforming mobility and affecting multiple industries. Radars are the critical sensor enabling autonomous driving and life-saving advanced driver-assistance system (ADAS) capabilities. Autonomous and ADAS vehicles equipped with radars have already been deployed on public roads. The primary radars' role is to provide high-accuracy information on the vehicle surrounding. However, radars are limited by their line-of-sight. This limitation is the most prominent in urban environments, where targets, such as vehicles and pedestrians, may be occluded by urban constructions. These ``around-the-corner'' hazardous objects can only be sensed via the non-line of sight (NLOS) propagation conditions. We propose to derive the breakthrough automotive radar NLOS-sensing capabilities employing neural network-based techniques. The main idea of the proposed approach is to exploit radar's high-resolution Multiple Input Multiple Output (MIMO) technology in the range-azimuth domain to obtain unique signatures at the radar receiver induced by reflections from targets, without a-prior knowledge of the walls' position and geometry. This study drives a radar signal received from the wall and NLOS multipath model of the target. The radar echo is processed using a pre-processing chain with EffienetNet b1 CNN to reveal the location of the wall, and algorithms are developed to localize the NLOS target in an urban scenario. The anticipated innovative NLOS sensing capabilities are expected to enable reliable autonomous and ADAS vehicles, thus having a significant commercial, economic, and societal impact.

ABSTRACT. Dual-function radar and communication (DFRC) is expected to significantly contribute to modern smart transportation applications. Next-generation intelligent vehicles are required to provide high-throughput communication services with high-performance detection and localization services at a centimeter-level resolution. The integration of radar and communication functions provides an opportunity for enhanced sensing and communication capabilities.

The main objective of this research is to maximize communication and sensing performance by DFRC waveform design. Optimal radar and communication integration requires efficient resource sharing, considering both spatial and spectral aspects. The available transmitter power needs to be efficiently allocated between the transmit antenna elements and spectrum. The main challenge in DFRC waveform design is the joint consideration of different optimization criteria needed for communication and sensing.

The main idea of the proposed approach is a cognitive waveform design in spatial and spectral domains , where the transmit waveform is dynamically adapted according to past observations. The radar performance is evaluated using the Bayesian Fisher Information for estimating the target parameters, such as direction-of-arrival (DOA) and range, where the posterior distribution is sequentially updated given past observations. This approach allows us to invest the power allocated for the radar function effectively and concentrate it only in the appropriate directions or sub-carriers. As a result, a significant amount of power is saved and can be allocated to enable high-capacity communication.

Three DFRC concepts are investigated in this work: 1. Radar-centric: maximizing radar estimation performance while satisfying minimal communication capacity requirements. 2. Communication-centric: maximizing the communication capacity while satisfying minimal radar estimation requirements. 3. Weighted criterion: considering both estimation accuracy and communication capacity performances with some pre-determined weights for radar and communications performance.

The performance of the proposed approach is evaluated in various practical scenarios, with various signal-to-noise ratio (SNR) values and multi-user communications architectures.

Our results show that the proposed approach efficiently allocates the power between the radar and communication and, thus, accurately estimates targets' parameters, while maintaining high-rate communication.

ABSTRACT. Abstract In the era of autonomous vehicles, optimizing and orchestrating the travel routes for all vehicles on roads is crucial to enhance traffic flow and efficiency. Travel Route Scheduling (TRS) aims to determine the shortest route (in terms of individual arrival time) to the destination while avoiding stops at intersections. Given the current global road state, TRS finds the shortest route for the next to-enter vehicle. This study assumes that vehicles are remotely controlled/instructed and are semi-autonomous or autonomous, with the (possibly distributed blockchain-based) control center constantly aware of their location, speed, and final destination. Under these assumptions, the algorithms presented in this work offer an optimal solution. The Algorithm The principal algorithm in this study is predicated on Dijkstra’s algorithm [1], where the node's weights undergo dynamic changes. Our algorithm, which incorporates dynamic weights along with its associated data structure and proof of correctness, may holds application within the scope of communication networks and other distributed algorithms involving multiple agents. The data structure of the algorithm is as follows: A directed graph representing the road network, where the junctions transform as nodes and the lanes as edges. Each edge is characterized by a length parameter, while each node possesses attributes such as length, conflicts function, list of vehicles movements, and wait time function. The theorem is as follows: The calculated arrival time for a vehicle to traverse from a source node to a designated node v is minimized. The objective of the algorithm is to find the shortest route for a particular vehicle that enters the system at a particular time. A comprehensive demonstration is available within the context of in [2]. Conclusion This study addresses the challenge of Travel Route Scheduling for vehicles on the road, focusing on eliminating stops at intersections. The underlying assumption is that these vehicles are connected via communication channels. The vehicles can be heterogeneous, autonomic, and/or human-driven, possibly instructed via a mobile application operated by a centralized or distributed (e.g., blockchain-driven) entity. It is noteworthy that while this algorithm was initially devised to address the challenge of Travel Route Scheduling without stops for vehicles on the road, its applicability extends to other domains, such as communication routing and multi-agent distributed systems. References [1] Edsger W Dijkstra. A note on two problems in connexion with graphs. Numerische mathematik, 1(1):269–271, 1959 . [2] Demonstration of the algorithm, 2024. https://drive.google.com/file/d/10imGkvc5ihDfIuIgYFPLON4LSu3-xzVg/view?usp=sharing

ABSTRACT. Video cameras are used to provide real-time feedback in automatic control systems, such as teleoperated driving systems. Raw video is coded, to meet network bandwidth, and finally transmitted to the human operator. In video coding standards (e.g. HEVC or H264) I-pictures tend to be greater than P and B frames, because temporal redundancy is not exploited at I-frames, only spatial redundancy is utilized. In ultra-low delay applications (e.g. teleoperated driving and cloud gaming) spikes in sizes of I-pictures may cause a significant increment in end-to-end (as well as glass-to-glass) delays, since it's required more time to transmit the I-picture. On the other hand I-frames are necessary in ultra-low latency video communication systems, since retransmission of lost or corrupted packets is not feasible in such applications. Therefore, I-frames are used to clean up visual degradations inflicted by lost or corrupted packets. We check three encoders on magnitudes of I-picture spikes: kvazaar, NVIDIA HW encoder (SDK 10) and x265, notice that kvazaar and x265 are SW encoders. The kvazaar encoder is not appropriate for ultra-low delay video communication systems since it generates I-pictures which are 5x-10x above the expected size and consequently significant glass-to-glass delay is expected. NVIDIA HW and x265 encoders produce moderate spikes at I-frame sizes (4x-5x times) and hence they are more suitable for very low delay video streaming than the kvazaar encoder.

ABSTRACT. Training autonomous vehicles to be both safe and deployable using Reinforcement Learning (RL) is an extremely challenging task. While RL is an attractive approach, traditional methods involving random exploration can lead to risky or unsafe behaviors. In practice, it is crucial for agents to prioritize safety while pursuing their objectives. The common approaches of associating safety violations with negative rewards or constraining agent behavior can be overly conservative and impractical, and one of the main limitations of learning the mistakes is obtaining samples where the ego vehicle ends up in a collision, even in a simulation, as collisions become increasingly rare as the driving policy improves. Our work addresses these challenges, and our concrete objective is to train autonomous vehicles to act safely without compromising their ability to find near-optimal behaviors, and without the need to spend dozens of hours driving in a simulator for every collision from which they learn to improve the behavior.

While our overall objective is to train autonomous vehicles to ensure zero collisions, mistakes are an inevitable part of any learning process. As a result, trying to ensure complete safety for agents before deployment is infeasible. Instead, we propose to embrace mistakes as a natural part of an agent’s existence, but to prevent the agent from ever repeating any catastrophic mistake [2]. This would, eventually, lead to zero or near-zero collisions while monotonically reducing the number of casualties.

Our work is based on the idea of training a shield, i.e., a safety neural network that learns to classify which actions are safe based on the agents’ real world experience. Initial work on this approach [1][2] showed great potential in reducing the number of collisions of autonomous vehicles. We propose to use the idea of Generative Adversarial Network (GAN) [3] for creating challenging problem instances for the agent and improving its performance and stability, by addressing the challenge of collecting unsafe samples.

We intend to demonstrate the effectiveness of the proposed approach in many driving scenarios (highway, merging, etc.) [4]. The approach is composed of three main components: An agent policy algorithm, such as Proximal Policy Optimization [5] which learns to select actions that maximize a predefined reward, A Shield; neural network for classifying actions to "safe" and "unsafe" from every state, And an adversarial network that attempts to generate problem instances aim to trick the shield. The examples provided by the generator help the shield to rapidly improve its predictions, without the need to wait for the challenging scenarios to naturally arise. Addressing various crucial issues, such as the optimal balance between these components, is also a significant aspect of this research.

We expect that over the course of training, the adversarial network will improve in creating challenging problem instances, as the shield network will improve in classifying actions that lead to collision correctly. In addition, we expect this novel approach to dramatically reduce the time required to train policies for autonomous vehicles and that the resulting policies would be much safer compared to the standard way of collecting data via environment interaction only.

References

1. Shahaf S. Shperberg, Bo Liu, Alessandro Allievi, Peter Stone. ”A RuleBased Shield: Accumulating Safety Rules from Catastrophic Action Effects.” In Proceedings of the 1st Conference on Lifelong Learning Agents, 2022.

2. Shahaf S. Shperberg, Bo Liu, and Peter Stone. Learning a shield from catastrophic action effects: Never repeat the same mistake, 2022.

3. Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. ”Generative Adversarial Networks.” arXiv preprint arXiv:1406.2661, 2014.

4. Edouard Leurent. An environment for autonomous driving decision-making. https://github.com/eleurent/highway-env, 2018.

5. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017.

ABSTRACT. Autonomous vehicles operate in dynamic and unpredictable environments, where rapid decision-making is critical to safety and efficiency. In such environments, the performance and safety of the vehicles can be highly affected by their collaboration, which depends on the ability of agents to share insights regarding the distinction between safe and unsafe actions. The complexity of multi-agent environments often necessitates that the agent’s perception of the environment may differ based on the sensory data it processes, such as visual and LiDAR data.

This research seeks to bridge these perceptual gaps by enabling reinforcement learning (RL) agents to learn collectively from shared experiences of safety-related incidents. RL is an attractive approach and widely studied; however, traditional methods for safety like linking safety violations to negative rewards or restricting agent actions tend to be overly cautious and often impractical, such as vehicles not moving at all to avoid collisions.

Our study introduces CoShieldPPO, this algorithm builds upon the concept of a safety mechanism known as shielding (Alshiekh et al., 2018), and the contribution of Shperberg et al. (2022). Shielding in RL helps prevent unsafe actions during training and deployment phases by masking out potentially hazardous decisions. At the heart of the CoShieldPPO algorithm is a cooperative shield component, designed to facilitate this knowledge exchange by serving as a safeguard used by all agents against unsafe actions. Using an encoder for each distinct observation type of an agent, the system learns a latent representation of observation that the shared shield can interpret. This latent observation space allows agents to recognize and avoid mistakes made by others. This shared learning approach is hypothesized to reduce catastrophic outcomes, leading to a safer policy. This methodology not only supports safer policy but also optimizes the learning process by focusing on significant safety breaches and avoiding inefficient exploration of unsafe actions. During training, agents contribute to a buffer with instances of both safe and unsafe actions. For N agents, the cooperative shield will be updated N times in each update step, and each encoder will be updated as the number of agents using this encoder observation type, accelerates the training process of the shield in comparison to a shield-per-agent approach. This update process also considers the specific type of observations each agent provides, whether it be image-based or LiDAR, to improve the latent representation of incoming observations.

Evaluation of our method will be conducted within a simulated environment for autonomous driving (Leurent, 2018). Key metrics for evaluation include the rate of unsafe actions, such as collisions, and the mean undiscounted episodic return, which together provide a comprehensive assessment of both safety and performance efficacy. We expect a substantial reduction in collisions during the early stages of deploying and training new autonomous vehicles. Furthermore, these fewer collisions are likely to lead to higher episodic returns, indicating better policies in an early phase. This aligns with the goals of advancing autonomous driving technologies and emphasizes the importance of collaborative learning in achieving safety enhancement within the realm of smart transportation.

References:

Alshiekh, Mohammed et al. (2018). “Safe reinforcement learning via shielding”. In: Proceedings of the AAAI conference on artificial intelligence. Vol. 32. 1.

Leurent, Edouard (May 2018). An Environment for Autonomous Driving Decision-Making. Version 1.4. URL: https://github.com/eleurent/highway-env.

Shperberg, Shahaf S et al. (2022). “Learning a Shield from Catastrophic Action Effects: Never Repeat the Same Mistake”. In: arXiv preprint arXiv:2202.09516.

ABSTRACT. Automotive radar is the main sensor enabling autonomous driving and active safety features. It is required to provide high-resolution information on the vehicle's surroundings, accurately localize surrounding objects, and estimate their velocity in two dimensions. Conventional automotive radars operating in the far-field regime estimate only the target's radial velocity and cannot obtain its tangential velocity. However, the near-field propagation conditions allow the tangential radar target velocity estimation. This work proposes to extend the radar aperture using the synthetic aperture radar (SAR) approach for automotive applications to extend the near-field operation conditions to cover the automotive radar ranges of interest. This work derives the near-field synthetic aperture model and defines the near-field synthetic aperture to conduct an identifiability study using the Cramér-Rao bound for the near-field model. It is demonstrated that it is possible to estimate the tangential radar target velocity in practical automotive scenarios.

ABSTRACT. The main objective of this work is to investigate the use of foundation models, such as CLIP [1] and Google/vit-base-patch16-224-in21k [2], for vision tasks that differ from the tasks used for their training. This work will investigate whether these models can be used as "frozen", i.e., with predefined features or need additional refinement. It will be further investigated whether these models could outperform conventional smaller models trained per specific datasets. The training speed and the required amount of training data, compared with the conventional task-specific models, will be investigated.

Specifically, this work focuses on vision tasks that include the estimation of image distortions. First, the vision tasks that estimate simple geometrical transformations between pairs of images will be considered. These tasks include simulations of applying geometrical affine transformations, rotation, and translation on an image and predicting the transformation parameters as a regression task, such as the rotated angle degree and the horizontal/vertical shifts. These tasks are usually used as pretext tasks for unsupervised training. The preliminary results demonstrated the vision transformers' possible efficiency in extracting features relevant to geometrical transformations. It is also shown that additional refinement of foundation models is needed.

Next, the more complicated image distortion-related tasks of predicting the dynamical fundamental matrix are considered. Estimating the fundamental matrix is a well-known problem in computer vision. The fundamental matrix describes the geometrical relationship between a pair of stereo-camera images of the same scene. This problem was intensively studied in the literature using conventional methods. For stereo cameras, the fundamental matrix is independent of the camera motion. Instead, we consider every two consecutive frames of the static scene as if the stereo camera generated them, and the fundamental matrix is estimated per timestamp. Another possibility is estimating the scene's rotation matrix and camera shift at each timestamp. It is proposed to perform the fundamental matrix prediction using the deep neural network with two modules. The first module is a Siamese-like extracting feature from two scene frames in parallel. The feature extraction is performed by refining a vision transformer, particularly the CLIP vision transformer. The second module is a regression multi-layer perceptron (MLP). The output of the MLP generates a 9-element vector, which is then manipulated to comply with the mathematical properties of the 3x3 fundamental matrix. The proposed approach will be extensively tested and compared with the state-of-the-art models. The same network architecture will be considered for other classification/regression tasks in computer vision.

Successful application of vision transformers for estimating the fundamental matrix will help improve computer vision tasks that rely on understanding the geometric relationship between images, such as 3D reconstruction, motion tracking, and navigation systems in autonomous vehicles. The findings and methodologies from this research are expected to contribute to the broader adoption of vision transformers in computer vision tasks.

ABSTRACT. The success of advanced driver assistance systems (ADAS) and autonomous vehicles (AV) largely depends on the sensing suit performance. The modern sensing suit contains cameras, LiDARs, and radars. Radars sense vehicle surroundings by analyzing electromagnetic echoes from reflecting surfaces. In urban scenarios, reflections from man-made structures, like buildings and guardrails, may induce multipath propagation phenomena. These indirect radar echoes often mix with direct reflections, creating confusing "ghost" targets that degrade radar performance. Conventional multipath phenomena modeling is challenging in dynamic urban scenarios due to the scene variability. Therefore, the conventional ray-tracing approaches introduced for cellular networks fail to provide an adequate solution for automotive radars. Data-driven deep neural network (DNN) approaches have been proposed to address this challenge, but they require extensive, accurately labeled datasets. Radar data labeling is cumbersome and nonintuitive compared to visual images. Therefore, radar data set labeling is highly challenging and expensive. To the best of our knowledge, the current datasets that provide ``ghost'' target labeling are either nonpublic, small, incomplete, contain significant labeling inaccuracies, are only coarse-labeled, and are missing multiple sensing modalities. This work introduces an automated labeling approach for automotive radar datasets, aiming to enhance publicly available large datasets with detailed labeling of "ghost" targets. The proposed approach leverages LiDAR measurements to estimate the geometry of reflective surfaces, utilizing this information to predict induced multipath effects and potential "ghost" targets. The main novelty of this work is the automated detailed multipath labeling approach for conventional large automotive radar datasets. This information is crucial for developing DNN-based automotive radar processing suitable for multipath-dominated urban environments. The proposed approach consists of three stages. First, the LiDAR measurements are used to estimate the multipath-generating reflectors' locations and their parameters. Next, the estimated reflectors are used to predict the induced multipath phenomenon and possible ``ghost'' targets. Lastly, this information is used to label the ``ghost'' targets within the original database and, thus, to generate the accurate and fine-granularity dataset of ``ghost'' targets. In addition to the ``ghost'' targets labeling, the proposed approach provides labeling of the actual reflectors. This information enables associating the ``ghost'' targets with the corresponding true objects. The performance of the proposed multipath labeling approach is evaluated using the radar ``Ghost Dataset'', with manually annotated multipath reflections in various automotive scenarios. The results demonstrated the proposed approach's high precision, recall, and F1-score for identifying and categorizing multipath reflections, types, and reflection orders. In addition, the proposed approach outperformed the LiDAR-based. Accurate radar multipath labeling capabilities of the proposed approach allow the introduction of multipath labels into large, publicly available radar datasets lacking the radar multipath labels. The availability of such datasets is expected to mitigate one of the significant challenges currently limiting the derivation of robust DNN-based automotive radar processing for multipath-dominated scenarios. The outcomes of this research expect substantial commercial, economic, and societal benefits by enabling AVs to operate reliably in complex urban environments.

ABSTRACT. Driver's interaction with a vehicle via automatic gesture recognition is expected to enhance driving safety by decreasing driver's distraction. Optical and infrared-based gesture recognition systems are limited by occlusions, poor lighting, and varying thermal conditions and, therefore, have limited performance in practical in-cabin applications. Radars are insensitive to lighting or thermal conditions and, therefore, are more suitable for in-cabin applications. However, the spatial resolution of conventional radars is insufficient for accurate gesture recognition. The main objective of this research is to derive an accurate gesture recognition approach using low-resolution radars. The main idea is to build a DL model that reconstructs high-resolution radar measurements and performs gesture recognition simultaneously. The ultimate high-resolution data adjusts for the gesture recognition task, enhancing its effectiveness compared with simple super-resolution. This allows for resource savings by avoiding the reconstruction of details that are irrelevant to gesture recognition, resulting in a more efficient model. The proposed approach combines conventional signal processing and deep learning methods. The radar echoes are arranged in 3D data cubes and processed using a super-resolution model to enhance range and Doppler resolution. The FFT is used to generate the range-Doppler maps, which enter the deep neural network for efficient gesture recognition. The preliminary results demonstrated the proposed approach's efficiency in achieving high gesture recognition performance using conventional low-resolution radars. Enhanced gesture recognition systems are expected to provide more intuitive and safer human-vehicle interactions by enabling more natural and reliable user inputs for controlling vehicle functions, such as navigation, infotainment, and climate control, without the need for physical contact. The gesture recognition's improved accuracy and robustness can greatly benefit advanced driver assistance systems (ADAS), contributing to higher levels of vehicle autonomy and better user experiences.

ABSTRACT. Remote driving serves as a viable solution in situations where fully autonomous vehicles encounter critical events, such as sensor failures. However, implementing remote driving poses certain technical challenges, including the need to ensure high-quality video transmission to the remote driver. Additionally, in scenarios involving poor road conditions, multiple autonomous vehicles may simultaneously require remote driving assistance at specific locations, straining the communication infrastructure.

To address these challenges, we propose a novel approach that involves compression of the driving video using a driving safety model. This model intelligently prioritizes key objects within the frame, resulting in improved compression quality. An initial experiment demonstrated that 60% of the required bitrate can be reduced while retaining 90% of the perceived quality.

ABSTRACT. Vision-based autonomous driving requires reliable and efficient object detection. This work proposes a DiffusionDet-based framework that exploits data fusion from the monocular camera and depth sensor to provide the RGB and depth (RGB-D) data. Within this framework, ground truth bounding boxes are randomly reshaped as part of the training phase, allowing the model to learn the reverse diffusion process of noise addition. The system methodically enhances a randomly generated set of boxes at the inference stage, guiding them toward accurate final detections. By integrating the textural and color features from RGB images with the spatial depth information from the LiDAR sensors, the proposed framework employs a feature fusion that substantially enhances object detection of automotive targets. The 2.3 AP gain in detecting automotive targets is achieved through comprehensive experiments using the KITTI dataset. Specifically, the improved performance of the proposed approach in detecting small objects is demonstrated.

ABSTRACT. The Light Rails sector is one of the sectors which can be considered according to the System of Systems (SoS) definition, since it includes the characteristics, attributes and behavior of SoS [1-2]. Our research develops G-QMS in Sectors of SoS which are characterized by a global and multi-organizational deployment (denotes as G-organization), while exploring the concepts, structure, contents and behavior of these G-QMS. Particularly, Light Rails sector is characterized by highly complex G-organizational structures. However, since it evolved from a traditional, experienced sector of the railway, these structures are proven and well-known. From the Quality Management System (QMS) perspective, SoS projects of this sector adopt a dual-system structure, initiated by the client through the contractual process that encompasses the supervision organizations and these of implementation. This dual structure inherently embeds controls across every facet and phase of the project, ensuring comprehensive oversight and quality assurance throughout the project. For every SoS project the G-organizational structure is established specifically as a special purpose G-organization, and it predominantly comprises organizations that are recognized experts in their respective fields and possess a global operational footprint. These G-organizations are involved in multiple projects worldwide, each managed by local organizational units dedicated to the specific contractual obligations of the project at hand. A local organizational unit typically holds the relevant professional management functions, including QMS. Due to these, the Light Rails sector is an important partner in the definition and development of G-QMS in Sectors of SoS. Our research initial objective was to create foundational principles for defining and modelling this kind of G-QMS with the unique characteristics and needs for the SoS [3]. The research methodology using structured qualitative research, based on semi-structured interviews and employing the Grounded Theory [4-5] combined with an analytical review and professional experience. The analytical review integrated rapidly evolving disciplines of QMS [6], SoS [7-8], Globalization [9] and Systems approaches chiefly Systems Thinking [10-11] and laid the foundation for this innovative field of research. From the literature it is found that each of these disciplines is generally still in a preliminary state in referring to organizational systems in particular those which are also global. Likewise, they emphasize the importance and potential contribution in the further progress of these disciplines in G-organizational applications. Our exploratory research exposed eight base anchors underlie any model of G-QMS in Sectors of SoS, moreover, a consolidated view of them can be considered as an initial model of G-QMS in Sectors of SoS [3]. Figure 1. illustrates a graphic depiction of the G-QMS in Sectors of SoS preliminary structure based on the consolidated view of the eight base anchors, indicating the following: [1-BA] The G-QMS model includes a headquarters entity (CORE) with an umbrella-view of the entire G-QMS, which is an aggregate model of QMSs chiefly the constituent systems’ existing QMS. The CORE situated on the top-level of command and control, it maintains the communication channels, which include a dual channel of control and reporting with each constituent system's QMS. [2-BA] The model includes the balance component which defines the level of dominance and centralization of the CORE in relationship to the independence level of the QMSs, and which varies between the QMSs in the same G-QMS, and thus, defines the model structure which contains a variety of structural configurations. The balance component also relates to the coordination factor in the base of the structural fabric creation of the G-QMS, a necessary condition for the G-QMS structure realization. [3-BA] Consequently, the aggregate model is a mosaic that does not strive for uniformity, but rather is capable of containing heterogeneity. [4-BA] The model defines a sorting factor for the SoS's constituent systems, by including an additional dimension of classification for the constituent systems' QMS.

Figure 1. G-QMS in Sectors of SoS - graphic depiction of preliminary structure (from Agmon et al., 2022, Figure 10, [3])

The model framework includes common aspects defined at the CORE level and shared by all, alongside the unique aspects of each constituent system's QMS that are determined internally. [5-BA] The model framework of G-QMS in Sectors of SoS shall be developed and extended from the ISO standards framework by adding the necessitating aspects. [6-BA] Its current management tools and methods shall also be expanded. [7-BA] Furthermore, in that development the Systems Thinking perspectives and principles should also be generalized. [8-BA] Lastly, the G-QMS in Sectors of SoS model must be adaptable and have a dynamic structure that yields a multitude of constantly-changing structures. Further to this exploratory study, an extensive field research that conducted in real-world global SoS organizations including these from the Light Rails sector, has validated, strengthened and expanded these foundations. The field research employs the same methodological methods, but applies them to a much larger scope of data, therefore involving data structures and content analysis techniques in a more complex manner. That research proposes a structural model for G-QMS in Sectors of SoS, that encompasses the varies structures of SoS projects and the G-organizations realizing them. Furthermore, one of its strengths is that it outlines the recommended model, which is vital as it directly impacts the success level of SoS projects and the effectiveness of the tailored G-QMS in these organizations.

ABSTRACT. Autonomous Mobility on Demand (AMoD) services offers numerous benefits, such as lower operating costs, due to reduced fuel and insurance costs and have no driver (Howard and Dai, 2014; Fraedrich and Lenz, 2014), which makes it extremely attractive for future development, thus, attracted a lot of attention in the literature. The AmoD services are becoming a reality and in the near future, in metropolitan cities like Tel Aviv, it is expected that such a service will attract more than 10% of daily trips demand (Nahmias-Biran et al., 2023). However, only a few studies successfully tested and evaluated a full AMoD service on a large and realistic network simulating real-world conditions.

AMoD services fulfill four main tasks: dispatching, routing, ridesharing, and rebalancing. Dispatching assigns vehicles to customers based on availability, proximity, and battery level. Routing optimizes routes for profitability, while ridesharing serves multiple riders with one vehicle, reducing energy use but complicating trip planning with multiple route calculations (Zardini et al., 2021). The rebalancing task involves repositioning empty vehicles to optimize responsiveness and serve future demand (Dai et al., 2021). It is especially important because AMoD systems experience imbalance when some areas have more demand than others (Pavone et al., 2012).

In this study we utilize a combined trio of simulation tools: (1) SimMobility demand prediction simulator, (2) Aimsun Next road network simulator, and (3) Aimsun Ride operator tool. The predicted demand for private vehicles and AmoD requests was done using SimMobility simulator for Tel Aviv futuristic metropolis in 2040, while this demand is executed using the Aimsun Next simulator. Demand-supply feedback is taking place so that travel times in the network are being updated and feed the demand repeatedly until convergence. Simulation outputs countian 1.2M routes of private cars on the large-scale network with an emphasis on their energy consumption. To create an efficient service framework for AMOD operation, we adopt a mathematical model of a Markov decision process (MDP). MDP allows us to optimize tasks such as pickup, rebalancing and charging under demand and energy consumption constraints.

The output of the MDP model is function which suggests the optimal action of a single vehicle in the AmoD fleet needs to perform (charging, pick-up, rebalancing) at a certain point in time. Finally, we design and execute an operator using the Ride tool that simulates the vehicles in the urban environment of Tel Aviv metropolis, along with other road users, performing battery and charging monitoring and sends the AMoD’s to tasks according to the optimal policy proposed by the MDP model. We compared this operator to two policy scenarios: (1) Rebalancing to the highest demand area after drop-offs , and (2) Self-decision rebalancing of the AMoD vehicle after drop-offs.

The model demonstrates energy savings, ranging from 80 to 132 kwh, equivalent to an additional travel distance of 615 to 1015 kilometers for a fleet of 100 AMoD's. This work provides valuable insights for operators policymakers, and urban planners seeking sustainable and optimized solutions for the integration of AMoD services in metropolitan environments.

ABSTRACT. Problem/Challenge Statement Shared dockless electric scooters (SDES) emerged as a new transportation mode in 2017 (Caulfield & Oeschger, 2020; POLIS, 2019). The suitability for short-to-medium range trips, and close to zero emissions, position them as a promising and sustainable urban mobility mode (Abouelela et al., 2023). However, since being introduced, the SDES has faced constant criticism regarding safety, public disturbance, and diversion riders from using public transport rather than private vehicles (Abouelela et al., 2023; Brown, 2021; Fearnley & Johnsson, 2023). We claim that the future integration of SDES as a part of urban transportation systems is defined by municipal policy that could maintain their advantages while mitigating associated conflicts. Objectives This study aims to: • Evaluate spatio-temporal (mis)match between SDES demand and supply. • Estimate the effect of cycling infrastructure on SDES use. • Propose policy guidelines for the effective management of the SDES system. Methodology We analyze three datasets that represent SDES use in Tel Aviv-Yafo (July 2023): • Trip characteristics (1M records) • E-scooters’ status changes (3M records). • Smartphone app activation and subsequent decision to ride (700K records) provided by the major SDES operator. Findings and Implications The analysis of scooters’ use reveals that: - The actual SDES trajectories are, on average, 20% longer than the shortest path between the rider’s origin and destination, but their coverage by cycling paths is 15% higher than that of the shortest paths and, due to the use of the cycling paths, the speed over the actual trajectories is 15% higher than that along the shortest path. - Parking duration of scooters that are parked closer to cycling infrastructure is shorter than those parked far from it. - The match between demand and supply varies in space. In some areas, the unused vehicles are accumulating, while in some the demand remains unsatisfied (Figure 1). - SDES users’ behavior is “fast and frugal”: After activating the application, users continue with the ride if the vacant scooter is at up to 100-150 m distance. In case several scooters are nearby, the users choose the closest one. Based on these findings, we propose policy guidelines that aim at matching the SDES demand to supply. The major points of this policy are (1) The establishment of parking cells all over the city at a distance of ~250 meters from each other, and (2) The periodic relocation of scooters that are not activated for long to the parking cells with unsatisfied demand.

ABSTRACT. Shared Demand-Responsive Transport (DRT), like Uber Pool, is a flexible component of the future Public Transport (PT) that may essentially improve the service for users whose standard PT alternatives are ineffective. The system that combines fixed bus lines and DRT services can be thus the next step of the future PT network [1]. Yet too many DRT vehicles will easily become a burden for the urban transportation system. We investigate the approach to balancing between fixed bus lines and DRT options by combining unsupervised learning and spatially explicit simulation. This presentation focuses on constructing bus lines that can substitute the emerging DRT flows. For this purpose, we simulate traffic in the synthetic city where the only PT service is provided by a fleet of DRT vehicles whose maximal capacity is 4 passengers, and the DRT dispatching algorithm prioritizes shared rides over those with a single passenger. We then analyze DRT passengers’ flows to extract clusters of travelers’ trajectories. The idea is to substitute cluster(s) that comprise many trajectories with a new PT fix route(s) that would serve these passengers. Our presentation focuses on the search for trajectories’ clusters. To recognize trajectories’ similarity, We propose a Weighted Cumulative Common Sub-Sequences Similarity measure (Weighted CCSS) that generalizes previously proposed CCSS [2] by assigning disproportionally high weights to sub-trajectories that include passengers’ origins and destinations. Then, the WCCSS measure is employed in an adjusted spectral clustering algorithm [3] which provides essentially better results comparing to the DBSCAN algorithm that is commonly used for clustering vehicle trajectories’ [4, 5]. Finally, the major clusters are turned into bus lines while a DRT service is preserved for the areas without access to these lines. The User Equilibrium of the proposed system is investigated with the simulation model. Our research contributes to a more efficient public transportation architecture, which will immediately contribute to the ISTRC vision: less casualties, less delays and minimize environmental harm.

[1] Saif, Muhammad Atiullah, Charalampos Sipetas, and Miloš Mladenović. "A phase-based perspective on urban demand responsive transport: A case study of Viavan pilot in Helsinki Capital Region." Case Studies on Transport Policy 15 (2024): 101123.‏ [2] Ma, Dongfang, et al. "Potential routes extraction for urban customized bus based on vehicle trajectory clustering." IEEE Transactions on Intelligent Transportation Systems (2023).‏ [3] Von Luxburg, Ulrike. "A tutorial on spectral clustering." Statistics and computing 17 (2007): 395-416.‏ [4] Kim, Jiwon, and Hani S. Mahmassani. "Spatial and temporal characterization of travel patterns in a traffic network using vehicle trajectories." Transportation Research Procedia 9 (2015): 164-184.‏ [5] Tang, Jinjun, et al. "Exploring urban travel patterns using density-based clustering with multi-attributes from large-scaled vehicle trajectories." Physica A: Statistical Mechanics and its Applications 561 (2021): 125301.‏

ABSTRACT. In recent years, deep neural networks (DNNs) have become a leading solution to many problems. In aerospace, aircraft manufacturers are exploring how deep-learning-based technologies could decrease the cognitive load on pilots, thus increasing the safety and operational efficiency of, e.g., airports.

One such attempt is Airbus’ Autonomous Taxi, Take-Off and Landing (ATTOL) project, in which Airbus has been testing the integration of vision-based functionality to autonomously assist pilots during various flight phases. For example, the taxi phase, which often creates an increased cognitive load on pilots, could benefit from object classification of potential threats on the runway. Indeed, a number of DNNs are being tested for this purpose within Airbus.

Using DNNs in safety-critical industries requires safety assurances. However, DNNs are known to be prone to various vulnerabilities, and are not covered by existing aeronautic software certification standards. One prominent issue is the sensitivity of DNNs to small input perturbations, which may cause a change in classification and lead to fatal errors [1]. For example, object classification DNNs mounted on an aircraft could misclassify an operator on the tarmac during the taxi phase, which in turn might lead to accidents (e.g, the Haneda airport incident). These issues are a major hindrance to the integration of DNNs in aeronautical products.

A promising approach for addressing this need for safety assurances is by formally verifying the correctness of DNNs. Formal verification either finds a perturbation of a given scale that leads to malfunction, or produces a mathematical guarantee of the absence of such perturbations [2].

Our work describes a case study that demonstrates the robustness analysis of a classifier of runway objects encountered during the taxi phase, under development at Airbus. We analyze the network’s robustness with respect to various scales of brightness, contrast, and noise.

In order to improve the speed of the verification process, we propose an incremental algorithm that decreases the number of verification queries. The algorithm arranges the queries monotonically, allowing a more effective scan that derives the robustness of the network for some scale of perturbations from its robustness for greater scale of perturbations.

In our experiments, we studied 1145 images correctly classified by the network. Our results suggest that the network is considerably more vulnerable to noise than to brightness or contrast perturbations, and that our incremental algorithm reduces the number of verifier calls by nearly 60%.

Moving forward, we aim to assess additional networks developed by Airbus. We hope that our work will serve as a milestone towards making formal verification a common practice in the aviation industry.

References:

[1] Goodfellow, I., Shlens, J., Szegedy, C.: Explaining and Harnessing Adversarial Examples (2014), technical report. http://arxiv.org/abs/1412.6572

[2] Katz, G., Barrett, C., Dill, D., Julian, K., Kochenderfer, M.: Reluplex: a Calculus for Reasoning about Deep Neural Networks. Formal Methods in System Design (FMSD) (2021)

ABSTRACT. Traffic congestion is a worldwide problem in modern societies. It can be broadly divided into recurrent (i.e. peak rush hours) and non-recurrent (i.e. caused by accidents or other hazards). The topic has long been researched, mainly by mathematical models that describe physical characteristics. The advances in data collection methods and data availability enable the use of data science methods. This paper strives to develop a method for detecting non-recurrent congestion in freeways by utilizing unsupervised machine learning methods. The developed capability will allow for flexibility and adaptive behavior to the irregular events occurring on the road network and causing the non-recurrent congestion. This behavior will be achieved by using anomaly and outlier detection methods based on unsupervised machine learning approaches with relevant data for the area of interest. Method: The ability to detect congestion build-up in a timely manner is crucial for effective traffic management and adequate response time for traffic mitigation. Many kinds of sensors are used for traffic-related measurements, most of which are statically placed along the freeways. These sensors provide input for both online applications and databases for offline processing, while the basic measured parameters are flow, occupancy, and speed. The proposed method uses these measured quantities to detect non-recurrent congestion while disregarding the recurrent ones. Each such cluster of sensors can be considered a source of spatiotemporal dataset with non-uniform spatial distribution and constant integration time of . The data is assumed to be with period of (e.g., 24 hours). Every one of these sets provide a set of historical data, while together they provide data sets, where the dimensions are the number of sensor clusters and number of considered days respectively. This matrix of datasets has been used for unsupervised training focused on detecting the non-recurrent congestion by using spatiotemporal anomaly detection. For this purpose, several anomaly detection algorithms were examined (a simple statistical approach, a few variations of robust covariance, one class SVM, and isolation tree). The ground rule is to apply the chosen method independently on each dataset. This approach has been tested on two databases. For initial testing a database of about a month of Ayalon highways was used. This database lacks incident recorded information, making evaluation of performance relative to ground truth nearly impossible. The second database that has been used is CalTrans PeMS SR73 North freeway recordings for 2016 as well as incident reports recorded in the system. The performance evaluation shows good results for detection rate, however false-alarm rate is lacking due to non-completeness of ground-truth data.

ABSTRACT. Traffic prediction is an essential component in Intelligent transportation systems (ITS) and it enables an understanding of the system that in turn allows for the utilization and optimization of transportation resources, infrastructure, and services. Graph-based learning has been utilized for many different traffic prediction tasks. However, to our knowledge, there has been no comparison of the different methods for building the adjacency matrix. In this extended abstract we present our research on comparing different adjacency matrix building methods for transportation networks. We compare three approaches for traffic prediction: road-based, distance-based, and similarity-based methods. We utilize a spatial-temporal graph convolution network (ST-GCN) and evaluate its performance on a traffic dataset, conducting experiments on a traffic flow prediction task. The findings suggest that creating implicit graph structures, which are automatically learned from data, may perform better than explicit graph structures that are manually defined based on prior knowledge.

ABSTRACT. The goal of signal timing coordination is to maximize throughput and minimize delay and stops. For under-saturated conditions, this goal is typically achieved by maximizing the duration of the coordinated phases. For over-saturated conditions, this goal is typically achieved by adjusting timing to minimize the adverse impact of queues [1]. Since environmental factors play an increasing role in traffic operations, an additional goal of timing coordination is to optimize the environmental performance of the traffic system. These include fuel consumption and a number of emission rates [2]. Progression optimization is a common approach for coordinating traffic signals along arterial streets as well as arterial networks. Several techniques have been developed for this purpose [3-4]. In recent years, another layer of sophistication has been added by enabling signal coordination that is tailored to individual traffic streams along paths between intersections [5]. The main control variable that affects coordination is the offset between the signals. This decision variable cannot optimize all measures simultaneously and the traffic engineer has to make tradeoffs among the various performance objectives.

We consider a case study consisting of two signalized intersections [call them INT- 6 and INT-7] along a major highway in the north of Israel and 12 connecting road sections. The principal control variable is the offset among the green phases along the main road. We employ the AIMSUN microscopic simulation software [6] to assess the performance of the traffic system, with special emphasis on environmental factors, for the following scenarios:

- Scenario 0: optimal cycle time for each signalized intersection (no coordination possible) - Scenario 1: the same cycle time for both intersections (signals are synchronized), progression priority at intersection 6 (i.e., INT-6 gets green light when head of platoon reaches it from INT-7). - Scenario 2: the same cycle time for both intersections (signals are synchronized), progression priority at intersection 7 (i.e., INT-7 gets green light when head of platoon reaches it from INT-6). - Scenario 3: the same cycle time for both intersections with maximum overlapping green (for different offsets).

The road network, traffic flow data, and the current traffic signal plans were coded into the simulation model. In addition, the flow characteristics and behavior of the vehicles in the model were calibrated to match reality in the field after a series of observations and assimilation of the driving behavior definitions of the simulation model. Different performance measures were evaluated for the four scenarios, including: delay time and number of stops, speed and travel time; fuel consumption, and VOC (volatile organic compounds), CO2,NOx, and PM (particulate matter). We derived performance results over one cycle time as a function of the offsets. When signals are synchronized, it is sufficient to illustrate one cycle time since the functions are periodic with the cycle time. While speed (and, consequently, travel time) achieve optimality (in opposite directions to each other) at a single offset within the cycle time, both PM and NOx attain two optima within the same cycle time. PM attains maximum when NOx is minimal, and vice versa. This illustrates the complexity of the optimization problem when environmental factors are considered.

The main conclusions from this study are: * Complex intersection systems can lead to multiple optima of the environmental performance measures. * Environmental performance measures are not consistent with mobility performance measures. * The traffic engineer has to determine the desired tradeoffs in the performance objectives.

References

[1] Texas Transportation Institute, TRAFFIC SIGNAL OPERATIONS HANDBOOK, Second Edition, FHWA 2011. [2] Shenzhen, D.; Xumei, C.; Lei, Y.; Xu, W. Arterial Offset Optimization Considering the Delay and Emission of Platoon: A Case Study in Beijing. Sustainability 2019, 11, 3882, 1-19. [3] Little, J. D. C., M. D. Kelson, and N. H. Gartner. MAXBAND: A Program for Setting Signals on Arterials and Triangular Networks. Transportation Research Record: Journal of the Transportation Research Board, 1981. 795: 40–46. [4] Gartner, N. H., S. F. Assmann, F. Lasaga, and D. L. Hou. MULTIBAND—A Variable-Bandwidth Arterial Progression Scheme. Transportation Research Record: Journal of the Transportation Research Board, 1990. 1287: 212–222. [5] Zatmeh-Kanj, S., Salomon, Y., Ben Chaim, M., and Gartner, N. Multi-Path Progression Optimization at an Urban Interchange: A Case Study.‏ ICASP 14, Dublin, Ireland, July 2023. [5] AIMSUN: Simulation and AI for Intelligent Mobility. https://www.aimsun.com/ (accessed May 1, 2024).

ABSTRACT. Emergency procedures for traffic-related events focus on blocking traffic, evaluating the event severity, and assisting evacuation. Less emphasis is given on informing road-users of the event extremity and guiding them how to behave. Road-users at the vicinity of a major event, who are not informed properly, can panic and react in unpredictable and odd ways, that can endanger themselves and others. Nowadays, technology enables road-operators to inform road-users via media, electronic boards, or designated emergency applications about traffic-related events and recommended behaviour. Less is known about the way road-users prefer to receive this information, or how to accustom it to their needs. Furthermore, in the era of shared road space of autonomous, connected vehicles, the issue of communication among those involved in an emergency event takes on new meanings. In the “transition period” when autonomous driving vehicles and human drivers will share the road space, communication needs may heighten. We aim to explore and model the recommended communication alternatives to improve road-operators and road-users’ interaction in emergency situations, assuming that the sources of information available to human drivers or passenger will differ from those of their automatic/ autonomous counterparts. This work will lead to recommendation for design requirements and guidelines.

ABSTRACT. Multiple consumer applications, such as communications and sensing, use electromagnetic waves, which increase human exposure to electromagnetic fields. High-frequency electromagnetic waves penetrate upper human skin layers, interacting with the capillary network and inducing nonthermal effects on the circulating red blood cells (RBC). This work developed an exposure system designed explicitly to investigate the effects of circulating blood exposure to plane electromagnetic waves. The developed system enables a systematic investigation of the electromagnetic field parameters that trigger macromolecule conformation. The system designs were meticulously optimized through extensive electromagnetic simulations. The modulated electromagnetic field (EMF) exposure was precisely applied at a frequency of 3.5 GHz, with an amplitude of 10dBm. The pulse depth was set at 4.0 μsec, and the rate was 8.0 μsec, all maintained consistently for 20 min. The EMF effect was verified using microwave dielectric measurements that were carried out in the frequency range from 500 MHz to 40 GHz. The analysis of the dielectric parameters of the main water relaxation peak was investigated. The effects of electromagnetic exposure, both at the molecular and cellular levels, were investigated using whole blood and blood hemolysate as the test systems. The fresh blood from eight healthy donors was collected for whole blood tests. Each blood sample was split into control and treatment portions (4 ml). The erythrocyte sedimentation rate (ESR) was measured according to standard protocol after 20 min of EMF exposure of whole blood in the exposed and control plastic Petri dishes. The elevated ESR was obtained in each EMF-exposed sample and compared with the control. The blood hemolysate was obtained from the whole blood by performing complete hemolysis (destroying) of the RBCs. The microwave dielectric spectrum for blood hemolysate was measured before (control) and after EMF exposure for 20 minutes. The control dielectric fitting parameters were compared with EMF-exposed solutions for five blood samples. The analysis shows that EMF exposure leads to a slight but statistically significant acceleration of water's dielectric relaxation, increasing dielectric strength. The observed effect may be associated with a change in the hemolysate balance between bounded and balk water. Our preliminary work demonstrates that exposure to EMF results in water state alternation in both the whole blood and blood hemolysate.

ABSTRACT. Hamas attack on October 7th created a crisis propagating to influence the central district of Israel. While war erupted in and the around Gaza strip, the civil population of Israel at large attempted to recover and shape some sort of normality. Their mobility requirements were adapted to the situation and kept changing. They were influenced by safety concerns and education system shutdown due to missiles threats, less workforce due to military reserve enlisting and more teleworking, no nightlife and less tourists. Concurrently, accessibility was influenced as well – as the public transportation (PT) services were disrupted due to transportation needs of the front. In the city of Tel Aviv, shared e-scooters offer on normal times a complementary transportation solution, delivered by 10,000 vehicles located throughout the city, which service almost 1M monthly trips (e.g., during October 2022). At time of crisis this capacity may offer agile addition to the disrupted transportation system. In this research we explore the changing mobility conditions and the role of e-scooters at this time of crisis. To examine the situation, we use e-scooter big-data from the Tel Aviv municipality Populus system, which provides us with Origin - Destination details as well as route choices; and the ticketing and bus trips information provided by the Ministry of Transportation “tikufim” database, through which PT patterns can be established, before and during the crisis. Using spatiotemporal and comparative analysis we explored the gradual adaptation of the city to the new circumstances. We examined the first 4 weeks of the war where the weekly 220K e-scooter trips were replaced by 60K, 90K, 100K and 120K trips respectively. At the same time the PT offerings fell by 60% and the validation rate dropped from over 2M per week to 800K, 1100K, 1480K, 1.6M respectively. We compare emerging spatial patterns to pre-crisis patterns – both to previous year and previous month, controlling for seasonal variations. The results show the initial drop in demand was also accompanied by changing centers of attraction for e-scooters, that seem to support both civil volunteering activities and military reserve efforts. The pre-existing temporal patterns of e-scooter use (and PT) which typically consisted of 2 peak daily demand, were replaced by a single peak pattern until reverting to more normal patterns. Throughout the period e-scooter trips’ characteristics remained similar if somewhat shorter and faster – around 2km/10:50 min for average trip. We further provide initial thoughts on harnessing the potential role of e-scooter at times of crisis.

ABSTRACT. Traffic capacity analysis is an essential part of transportation engineering and enables traffic planners to help design, manage and understand road dynamics. In this study, we propose a novel approach for defining and optimizing road capacity by combining both deep learning models and genetic algorithms. We conducted the research using data from the California PeMS system. The first part of the process is a deep learning model, which we built using a LSTM network. This model predicts the traffic speed given the traffic flow for the whole of the network. The second part is a genetic algorithm that utilizes the speed prediction model and aims to optimize network-wide traffic flow dynamics by trying to find an efficient combination of traffic flow for the whole network.