ABSTRACT. We investigate the effect of errors in collective robotic construction (CRC) on both construction time and the probability of correctly completing a specified structure. We ground our investigation in the TERMES distributed construction system, which uses local sensing and stigmergic rules that enable robots to navigate and build 3D structures. We perform an in depth analysis and categorization of action failures in CRC systems. We present an approach to mitigating action failures and preventing errors that prohibit completion of a structure by adding predictive local checks. We show that the predictive local checks can increase the probability of success by orders of magnitude in large structures. This work demonstrates the need to consider both construction time and the effect of errors in collective robotic construction.

ABSTRACT. Perimeter defense is an active area of research with a wide variety of applications. In perimeter defense, a team of defenders seeks to intercept a team of intruders before they reach the perimeter. Though the single defender case is relatively well studied, with multiple defenders significant complexity is introduced due to the added coordination and routing requirements for multi-agent systems. In this work, we present a conversion from the perimeter defense problem to an instance of the min-cost-max-flow problem for flow networks, and leverage existing efficient algorithms for network flows to solve both the task assignment and routing problems for perimeter defense concurrently. When considering homogeneous defender robots, the computed solution is optimal for any individual timestep. Additionally, we detail a deconflict-based strategy for dealing with heterogeneous defenders, and show in simulation that the proposed solutions match or outperform a naive greedy baseline.

ABSTRACT. We propose a classification-aware path planning architecture for a team of robots in order to traverse along the most informative paths with the objective of completing map classification tasks using localized (partial) observations from the environment. In this method, the neural network layers with parallel structure utilize each agent's memorized history and solve the path planning problem to achieve classification. The objective is to avoid visiting less informative regions and significantly reduce the total energy cost (e.g., battery life) when solving the classification problem. Moreover, the parallel design of the path planning structure reduces the training complexity drastically. The efficacy of our approach has been validated by a map classification problem in the simulation environment of satellite campus maps using quadcopters with on-board cameras.

ABSTRACT. In this paper, we present an algorithm that automatically encodes a user-defined complex 2D shape to a set of cells on a grid each characterizing a robot currently in the swarm. The algorithm is validated via up to 100 simulated robots as well as up to 100 physical robots. The results show that the goal configurations generated by the algorithm for the swarms with any size are consistent with the input shapes, moreover, it allows the swarm to adapt to the swarm size change quickly and robustly. A summary video of the presented algorithm can be found at: https://northwestern.box.com/s/94i2urop3psz7j1jqahxhhy8t78o6ggb

ABSTRACT. The prevalence of autonomous agents has raised the need for agents to cooperate without having the ability to coordinate their moves in advance, or communicate explicitly. This is referred to as ad-hoc teamwork. Prior work in this field has examined the possibility of leading a flock of simple, swarm-like, agents to a desired behavior that maximizes joint group utility, using informed agents that act within the flock. In this work we examine the problem of leading a flock of agents using signals. In this problem, the leading agents are equipped with a tool allowing them to send a simple signal to the flock, "calling" them to act in a desired way. However, the agents may misinterpret the signal with some probability, and head to the opposite direction. We examine the best behavior for a leading agent, that is, deciding when to signal, which depends on the signaling range of the leader, the probability of the signal misinterpretation, the sensing range of the flocking agents, and their current behavior. We extend the analysis to multiple leading agents, and show that their location within the flock also plays a role in the outcome. Finally, we examine the use of signals by leading a swarm of agents in the dispersion problem, which demonstrates the limitation of signals. Specifically, we show that signals may have no influence on the performance of the swarm, even if they are perfectly interpreted.

ABSTRACT. Lifelong and online Multi-Agent Pickup and Delivery is a task and path planning problem in which tasks arrive over time. Real-world applications may require decentralized solutions that do not currently exist. This work proposes a decentralized and Byzantine fault tolerant algorithm using blockchain that is competitive against current distributed task and path planning algorithms. At every timestep agents can query the blockchain to receive their best available task pairing and propose a transaction that contains their planned path. This transaction is voted upon by the blockchain network nodes and is stored in the replicated state across all nodes or is rejected, forcing the agent to re-plan. We demonstrate our approach in simulation, showing that it gains the decentralized Byzantine fault tolerant consensus from the blockchain while remaining competitive against current solutions in its makespan and service time.

ABSTRACT. We consider the problem of decentralized multi-robot kinodynamic motion planning in dynamic workspaces. The proposed approach leverages offline precomputation on an invariant planning representation (invariant geometric tree) for low latency online planning and replanning amidst unpredictably moving dynamic obstacles to generate kinodynamically feasible and collision-free time-parameterized polynomial trajectories. Simulation results with up to 10 robots in dynamic workspaces composed of varying obstacle densities (up to 30 % by volume) and speeds (up to 2.5 m/s) suggest the use of the proposed methodology for real-time kinodynamic replanning in dynamic workspaces.

ABSTRACT. In this paper we propose a method for optimizing the design of a roadmap, used for motion coordination of groups of automated guided vehicles for industrial environments. Considering the desired flows among different locations in the environment, we model the problem as a multi-commodity concurrent flow problem, which allows us to assign the directions of the paths in an optimized manner. The proposed solution is validated by means of simulations, exploiting realistic layouts, and comparing the performance of the system with those achieved with a baseline roadmap.

ABSTRACT. Moving a module in a modular robot is a very complex and error-prone process. Unlike in swarm, in the modular robots we are targeting, the moving module must keep the connection to, at least, one other module. In order to miniaturize each module to few millimeters, we have proposed a design which is using electrostatic actuator. However, this movement is composed of several attachment, detachment creating the movement and each small step can fail causing a module to break the connection. The idea developed in this paper consists in creating a new kind of deformable module allowing a movement which keeps the connection between the moving and the fixed modules. We detail the geometry and the practical constraints during the conception of this new module. We then validate the capability of motion for a module in an existing configuration. This implies the cooperation of some of the other modules placed along the path and we show in simulations that it exists a motion process to reach every free positions of the surface for a given configuration.

ABSTRACT. In decentralised autonomous systems it is the interactions between individual agents which govern the collective behaviours of the system. These local-level interactions are themselves often governed by an underlying network structure. These networks are particularly important for collective learning and decision-making whereby agents must gather evidence from their environment and propagate this information to other agents in the system. Models for collective behaviours may often rely upon the assumption of total connectivity between agents to provide effective information sharing within the system, but this assumption may be ill-advised. In this paper we investigate the impact that the underlying network has on performance in the context of collective learning. Through simulations we study small-world networks with varying levels of connectivity and randomness and conclude that totally-connected networks result in sub-optimal performance when compared to networks with less connectivity. Furthermore, we show that networks of high regularity outperform networks with increasing levels of random connectivity.

ABSTRACT. Redundancy and parallelism make decentralized multi-robot systems appealing solutions for the exploration of extreme environments. However, effective cooperation can require team-wide connectivity and a carefully designed communication. Several recently proposed decentralized connectivity maintenance approaches exploit elegant algebraic results drawn from spectral graph theory. Yet, these proposals are rarely taken beyond simulations or laboratory implementations. The contribution of this work is two-fold: (i) we describe the full-stack implementation---from hardware to software---of a decentralized control law for robust connectivity maintenance; and (ii) we assess, in the field, our robots' ability to correctly exchange the information required to execute it.

ABSTRACT. Robotics research needs complex hardware and software that is why simulation is often view as an alternative for testing. Large scale self-reconfiguring modular robotic systems needs a scalable simulation environment which cannot be physics-based. This paper presents VisibleSim, an open-source behavioral simulator for lattice-based modular robots that uses discrete-event simulation to simulate ensembles of up to millions of modules. We describe the principles behind the simulator and introduce its features and usage from a user standpoint. VisibleSim is built with extensibility, versatility, and flexibility in mind, can be used as a powerful visualization tool, and already has a proven track record with several modular robotic architectures.

ABSTRACT. Whilst swarms have potential in a range of applications, in practical real-world situations, we need easy ways to supervise and change the behaviour of the swarm to promote robust performance. In this paper, we design artificial supervision of swarms to enable an agent to interact with a swarm of robots and command it to efficiently search complex partially known environments. This is implemented through artificial evolution of human readable behaviour trees which represent supervisory strategies. In search and rescue (SAR) problems, considering uncertainty is crucial to achieve reliable performance. Therefore, we task supervisors to explore two complex environments subject to varying blockages which greatly hinder accessibility. We demonstrate the improved performance achieved with the evolved supervisors and produce robust search solutions which adapt to the uncertain conditions.

ABSTRACT. We present a method based on covariance intersection for cooperative localization with pairwise range-only relative measurements. Our method was designed for underwater robots equipped with an acoustic communication and ranging system. Range measurements are not sufficient to compute a complete relative $3$D position. Therefore, covariance intersection is performed in a transformed space along their relative estimated positions, while preserving cross-correlations between other state variables. Given the characteristics of the acoustic channel, only one robot can transmit data or a ranging request at a time, hence the pairwise limitation. We also present a heuristic for choosing a peer robot for a range measurement by maximizing mutual information. Our method places no further restrictions on the order, timing or scheduling of relative measurements. We evaluated our method for accuracy and consistency, and present results from simulations as well as outdoor experiments.

ABSTRACT. This paper presents a thorough experimental evaluation of an extended Gaussian Mixture Probability Hypothesis Density filter which is able to provide state estimates for the maintenance of a multi-robot formation, even when the communication fails and the tracking data are insufficient for maintaining a stable formation. The filter incorporates, firstly, absolute poses exchanged by the robots, and secondly, the geometry of the desired formation. By combining communicated data, information about the formation, and sensory detections, the resulting algorithm preserves accuracy in the state estimates despite frequent occurrences of long-duration sensing occlusions, and provides the necessary state information when the communication is sporadic or suffers from short-term outage. Differently from our previous contributions, in which the tracking strategy has only been tested in simulation, in this paper we present the results of experiments with a real multi-robot system. The results confirm that the algorithm enables robust formation maintenance in cluttered environments, under conditions affected by sporadic communication and high measurement uncertainty.

ABSTRACT. We study the problem of information sampling with a group of mobile robots from an unknown environment. Each robot is given a unique region in the environment for the sampling task. The objective of the robots is to visit a subset of locations in the environment such that the collected information is maximized, and consequently, the underlying information model matches as close to reality as possible. The robots have limited communication ranges, and therefore can only communicate when nearby one another. The robots operate in a stochastic environment and their control uncertainty is handled using factored Decentralized Markov Decision Processes (Dec-MDP). When two or more robots communicate, they share their past noisy observations and use a Gaussian mixture model to update their local information models. This in turn helps them to obtain a better Dec-MDP policy. Simulation results show that our proposed strategy is able to predict the information model closer to the ground truth version than compared to other algorithms. Furthermore, the reduction in the overall uncertainty is more than comparable algorithms.

ABSTRACT. In this paper, we present a Probability Hypothesis Density (PHD) filter based relative localization system for robotic swarms. The system is designed to use only local information collected by onboard lidar and camera sensors to identify and track other swarm members within proximity. The multi-sensor setup of the system accounts for the inability of single sensors to provide enough information for the simultaneous identification of teammates and estimation of their position. However, it also requires the implementation of sensor fusion techniques that do not employ complex computer vision or recognition algorithms, due to robots' limited computational capabilities. The use of the PHD filter is fostered by its inherent multi-sensor setup. Moreover, it aligns well with the overall goal of this localization system and swarm setup that does not require the association of a unique identifier to each team member. The system was tested on a team of four robots.

ABSTRACT. As our contribution to the effort of developing methods to make robots more adaptive and robust to dynamic environments, we have proposed our method of 'minimal surprise' in a series of previous works. In a multi-robot setting, we use evolutionary computation to evolve pairs of artificial neural networks: an actor network to select motor speeds and a predictor network to predict future sensor input. By rewarding for prediction accuracy, we give robots an innate, task-independent motivation to behave in structured and thus, predictable ways. While we previously focused on feasibility studies using abstract simulations, we now present our first results using realistic robot simulations and first experiments with real robot hardware. In a centralized online and onboard evolution approach, we show that minimize surprise works effectively on Thymio II robots in an area cleaning scenario.

ABSTRACT. Cooperative transportation is one of the essential tasks for multi-robot systems to imitate decentralized systems of social insects. However, in a situation with an obstacle on the pathway, multiple robots need to realize transportation and obstacle removal simultaneously. To address this multi-tasking problem, we first introduce a learning scenario and train robots' decentralized policies via multi-agent reinforcement learning. Next, we propose two virtual experiments with blindfold teams and homogeneous teams to analyze the individual behaviors of the trained robots. The results showed that three robots with different policies performed two tasks simultaneously as a team. One robot's policy tended to perform obstacle removal, and the other robots' policies tended to perform cooperative transportation. Further, the first robot's policy had the potential to perform two tasks simultaneously depending on the situation. Finally, we demonstrated the trained policies with three ground robots to show the feasibility of the system.

ABSTRACT. The Defense Advanced Research Projects Agency's (DARPA's) OFFensive Swarm-Enabled Tactics (OFFSET) program aims to develop a system architecture and algorithms for conducting urban missions with heterogeneous spatial swarms of up to 250 unmanned air and ground vehicles. Swarms' scale logistically prohibits modeling and tracking individual batteries, while highly variable battery lives make it difficult to determine, without such modeling, whether a vehicle has sufficient power to complete its tasks. The Swap algorithm manages battery variability by autonomously exchanging swarm vehicles with depleted batteries for ones with fresh batteries. Simulation-based evaluation demonstrates that Swap substantially increases mean task completion and reduces variance, thus increasing mission success and outcome consistency.

ABSTRACT. This paper develops a multi-agent heterogeneous search approach that leverages the sensing and motion capabilities of different agents to improve search performance (i.e., decrease search time and increase coverage efficiency). To do so, we build upon recent results in ergodic coverage methods for homogeneous teams, where the search paths of the agents are optimized so they spend time in regions proportionate to the expected likelihood of finding targets, while still covering the whole domain, thus balancing exploration and exploitation. This paper introduces a new method to extend ergodic coverage to teams of heterogeneous agents with varied sensing and motion capabilities. Specifically, we investigate methods of leveraging the spectral decomposition of a target information distribution to efficiently assign available agents to different regions of the domain and best match the agents' capabilities to the scale at which information needs to be searched for in these regions. Our numerical results show that distributing and assigning coverage responsibilities to agents based on their dynamic sensing capabilities leads to approximately 40% improvement with regard to a standard coverage metric (ergodicity) and a 15% improvement in time to search over a baseline approach that jointly plans search paths for all agents, averaged over 500 randomized experiments.

ABSTRACT. This paper studies a sequential adversarial incomplete information game, the attack-defense game, with multiple defenders against one attacker. The attacker has limited information on game configurations and makes guesses of the correct configuration based on observations of defenders' actions. Challenges for multi-agent incomplete information games include scalability in terms of agents' joint state and action space, and high dimensionality due to sequential actions. We tackle this problem by introducing deceptive actions for the defenders to mislead the attacker's belief of correct game configuration. We propose a k-step deception strategy for the defender team that forward simulates the attacker and defenders' actions within k steps and computes the locally optimal action. We present results based on comparisons of different parameters in our deceptive strategy. Experiments show that our approach outperforms Bayesian Nash Equilibrium strategy, a strategy commonly used for adversarial incomplete information games, with higher expected rewards and less computation time.

ABSTRACT. We consider the problem of deploying robots to observe the evolution of a stochastic process in order to output a sequence of observations that fit some given specification. This problem often arises in contexts such as event reporting, situation depiction, and automated narrative generation. The paper extends our prior work by formulating and examining the multi-robot case: a team of robots move about, each recording what they observe, and, if they manage to capture some event, communicating that fact with the group. In the end, all events from all the robots are collated to provide a cumulative output. A plan is used to decide what each robot will attempt to capture next, based on the state of the world and the events that have been captured (collectively) so far. This paper focuses on the question of how to compute effective multi-robot plans. A monolithic treatment, involving the optimal selection of joint choices, i.e., choosing the next elements to attempt to capture by all robots, is formulated where costs are minimized in an expected sense. Since such plans are prohibitive to compute, variants based on an approximation scheme based on solving a sequence of individual planning problems are then introduced. This scheme sacrifices some solution quality but requires far less computational expense; we show this permits one to scale to greater numbers of robots.

ABSTRACT. Swarm robotics technology will be ready to move to the real world soon. However, its public perception is generally negatively impacted by science fiction. To improve its understanding and public trust, it is important to engage the public. In this paper we propose the use of an innovative, educational escape room to address this. Here we show participants increased their knowledge of swarm robotics through the game. Furthermore, we show Swarm Escape! was the catalyst for unsure players to adopt a positive attitude towards robot swarms. A discussion-style session was also organised to engage in a dialogue about this technology. We report concerns and key actions for the future given by participants. With this paper we hope to inspire swarm robotics researchers to embrace the culture of ethical governance and responsible research and innovation to realise its economic and societal benefits by means of research for society.

ABSTRACT. Swarm robotic systems aim to coordinate multiple robots flexibly in various environments and solve flexibly distributed cooperative tasks scattered in the environment through their coordination. To realize this work efficiently, self-organization for robot swarms suitable for the scale, shape, and structure of the task is essential. In this regard, this study considers the natural phenomena of crystal growth as the task for adaptive self-organization and proposes a self- organizing method using the phase-field method and cellular automata to represent crystal growth. The phase-field method and cellular automata are applied to the behavioral control transition of the robot from the environment search to surface search and the pattern design for constructing a target shape, respectively. Moreover, this study designed rulesets to form a similar shape focusing on the process of adhesive and layer growth and confirmed the self- organization of each of them. Finally, we discussed how each ruleset affects the asynchronous self-organization on the process of organizational formation.

ABSTRACT. Multimodal optimization is a very difficult task to search for all optimal solutions at once in optimization problems with multiple optimal solutions. Speciation using a proximity graph has been proposed to solve multimodal optimization problems. Gabriel graph (GG) and relative neighborhood graph (RNG) are often used as the proximity graph. The search efficiency is good when GG is used, but the discovery rate of the optimum solutions is lower than when RNG is used. In this study, we propose a new proximity graph with a parameter β named “β relaxed relative neighborhood graph” (βRNG) that can be generated relatively fast and has intermediate properties between GG (β=1) and RNG (β=2). Also, βRNG is adopted in SPSO-G (Speciation-based Particle Swarm Optimization using Graphs) for graph-based speciation. The performance of the proposed method is shown by optimizing well-known benchmark problems for “CEC’2013 special session and competition on niching methods for multimodal function optimization”.

ABSTRACT. This paper presents a control strategy for survivor searching in a dynamically changing flood zone by using a group of unmanned aerial vehicles (UAVs). Assuming that there are multiple groups of the survivors, the positions of which are time-varying and cannot be accurately located, the control strategy requires the UAVs to optimally cover possible locations of survivors in the flood zone. A robust adaptive controller has been proposed to implement the strategy, the feasibility of which is verified under simulations in the presence of time-varying uncertainties.

ABSTRACT. The consensus problem, which examines the possibility of reaching an agreement, is a fundamental problem in swarm robotics. In this paper, we examine a new variant of the problem, which concerns the ability to influence a swarm into reaching a desired consensus, focusing on leading a flock of agents to a desired orientation. We suggest to influence the flock by the use of an external agent, referred to as influencing agent, and examine its power over a flock with either a fixed topology or a switching topology. We prove that even a single influencing agent with a simple Face Desired Orientation behaviour that is injected into a static flock is sufficient for guaranteeing desired consensus of flocking agents which follow a Vicsek-inspired Model. When applying this concept in dynamic settings, we discuss the arising limitations, as well as restrictive conditions under which such influencing agents can force the entire flock to attain the desired consensus.

ABSTRACT. This paper aims to derive control laws for formation tracking and searching of multiple quadrotors. First, quadrotor dynamics and the network structure are defined. Second, the mode determination, whether a quadrotor should track or search in the next iteration, protocols are provided through utilizing the idea of Markov Decision Process(MDP).Third, the control laws for formation tracking are derived based on the consensus algorithm with the flexibility of the size of network confirmed. Forth, the control laws for searching three-dimensional area are provided. Searching control laws' formulas contain the reference points whose heights and coordinates on the horizontal planes are calculated respectively, which reduces the computational burden and the irrational or impractical travels of quadrotors. Finally, some simulations are shown to validate the effectiveness.

ABSTRACT. In this paper, a model-based-event triggered formation control algorithm for a multi-agent system is proposed. An essential task in designing a distributed control scheme for a Multi-Agent System should focus on the desired control performance and save communication and computation resources. At event-triggered control, the timing of performed instants is determined by predefined triggering, which is linked to system measurements. That means if the event triggering condition is violated, in the form of the current system measurements exceeding a certain threshold, the event trigger will occur. Compared to the time-triggered scheme, the event-triggered method can help reduce communication and calculation. At the end of this paper, simulation results will be shown to verify the proposed method.

ABSTRACT. We present a method to register individual members of a robotic swarm in an augmented reality (AR) display while showing relevant information about swarm dynamics to the user that would be otherwise hidden. In order to register the physical robots with the AR display, individual swarm members and clusters of the same group are identified by their color, and by blinking at a specific time interval that is distinct from the time interval at which their neighbors blink. We show that selecting an appropriate sequence is an instance of the distributed graph coloring problem, which can be solved in O(log (n)) time with n being the number of robots involved. We demonstrate our approach using a swarm chemistry simulation in which robots simulate individual atoms that form molecules following the rules of chemistry. An AR display is then used to display information about the internal state of individual swarm members as well as their topological relationship, corresponding to molecular bonds in a context that uses robot swarms to teach basic chemistry concepts.

ABSTRACT. Low stiffness, large stroke and relatively large axial force make Extensile Artificial Muscles (EFAMs) a great choice for continuum soft robots. EFAMs can be used as a soft structure of a robot that provides intrinsic actuation. Although FAM axial properties have been well studied, little is known about their flexural behavior. Static and dynamic flexural behavior of a cantilevered EFAM specimen was investigated over a pressure range of 5-100 psi with 5 psi increments. Based on the experimental data, static and dynamic properties were estimated with use of linear and non-linear beam models. The test method also provided dynamic response data that was used for determination of dynamic bending properties such as damping ratio and natural frequency. Experimentally-validated flexural properties of EFAMs are significant for accurate modeling of EFAM-driven continuum soft robots as well as structural and control considerations of the robots.

ABSTRACT. In this paper, we focus on an aerial surveillance system composed of single sentinel and multiple searchers. In the system, unidentified flight vehicles, UFVs, are surveillance targets. The searchers are required to approach the UFVs for identification. In this regard, however, faster UFVs move while avoiding the searchers. In order for the searchers to approach and detect such sophisticated UFVs, cooperative actions play an important role. Therefore, we apply reinforcement learning to the searchers. In the initial learning phase, however, searchers based on random actions seldom detect UFVs. Thus the searchers cannot acquire effective actions to approach the target UFV. Therefore, we further apply transfer learning to this problem. Through surveillance simulations, we show the effectiveness of the cooperative actions by the multiple searchers for the UFVs in the aerial surveillance system.

ABSTRACT. Today, mobile robot systems are designed for open public environments and can benefit from modeling uncertainties in their decision processes. For this reason, approaches based on Markov Decision Processes (MDP) are proposed to control autonomous robots since few decades. Optimally solving an MDP based model to coordinate a fleet of robots can rapidly become intractable. This paper focuses on the distributive property of Multi-agent Markov Decision Processes (MMDP) and identify an MMDP model as ``Completly Distributable'', ``Partially Distributable'' or ``Distributable with Coordination''. Based on those definitions, the classical multi-mobile-robot problems addressed in the literature are classified and discussed accordingly regarding the potential to compute an optimal solution in a distributive way.

ABSTRACT. Dynamic systems are usually described by differential equations, but formulating these equations requires a high level of expertise and a detailed understanding of the observed system to be modelled. In this work we present a data driven approach, which tries to find a parameterization of neural differential equations system to describe the underlying dynamic of the observed data. The presented method is applied to a multi-agent system with thousand agents.

ABSTRACT. Robot swarms have the potential to be used as an out-ofthe-box solution for storage and retrieval that is low cost, scalable to the needs of the task, and would require minimal set up and training for the users. We show in simulation that stochastic strategies based on random walk and probabilistic sampling of local boxes could give rise to competitive solutions to retrieve boxes and deliver them unordered, or following a predetermined order, within a storage scenario.

ABSTRACT. In this paper we study the multi-agent pursuit-evasion problem, and present an extension of the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) deep reinforcement learning algorithm. Previous pursuit-evasion advancements with MADDPG have focused on training capture strategies dependent on the restriction of evader movement with environmental features. We demonstrate a method to train pursuer agents to collaboratively surround and encircle an evader for reliable capture without a strategy rooted in environment entrapment (i.e. cornering). Our method utilizes a novel two-stage, variable-aggression, continuous reward function based on geometrical inscribed-circles (incircles), along with a corresponding observation space, with agents operating in a entrapment-disadvantaged environment. Our results show reliable capture of an intelligent, superior evader by a swarm of three trained pursuers in open space with our encircling strategy.

ABSTRACT. This paper presents a distributed multi-robot printing method which utilizes an optimization approach to decompose and allocate a printing task to a group of mobile robots. The motivation for this problem is to minimize the printing time of the robots by using an appropriate task decomposition algorithm. We present one such algorithm which decomposes an image into rasterized geodesic cells before allocating them to the robots for printing. In addition to this, we also present the design of a numerically controlled holonomic robot capable of spraying ink on smooth surfaces. Further, we use this robot to experimentally verify the results of this paper.

ABSTRACT. Resilient collective agent systems must react appropriately to unanticipated disturbances during runtime in order to satisfy the collective's goal. Collectives are characterized by emergent behaviors in which interactions between individual agents and the environment result in complex collective dynamics. Understanding how a disturbance to individual agents or to a collective's operational environment affect the collective's overall task performance is critical to designing adaptive behaviors for resilient collectives. This manuscript presents results for two disturbance types applied to a collective's best-of-N process: an addition that increases the collective's size, and an ablation that removes a site from the environment. The results show that the disturbance's timing is most impactful to the collective’s selection accuracy when the highest valued site is at an intermediate distance from the hub relative to other options, but the disturbance’s effect was subsumed by the effects of environmental bias for relatively easy or difficult problems.

ABSTRACT. A robotic swarm is a particular type of Multiagent System that employs a large number of simpler agents in order to cooperatively perform different tasks. Oftentimes, the implementation of complex swarm behaviors is a challenging task, and researchers have started to rely on machine learning techniques, which normally require large and complex training setups. In this paper, we explore a segregated navigation task, in which different groups of robots should navigate in a shared environment without mixing with others. Specially, we investigate if a robot trained in simpler scenarios, using a smaller number of robots and groups, can use the learned behavior in more complex scenarios. We performed a series of simulations varying the number of robots and groups and discuss that learning in simpler scenarios can be effective in the segregated navigation task.

ABSTRACT. In this paper, we present a controller for collision-free velocity tracking of a moving ground target by multiple unmanned aerial vehicles (UAVs). The controller combines a feedforward proportional-derivative (PD) control term and a term that is based on the gradient of an artificial potential function. We use Lasalle's invariance principle to analytically prove the convergence of the UAVs to a fixed formation above the target that tracks the target's velocity and provide mathematical guarantees on the UAVs' collision avoidance. As a result, the Euclidean distance between each pair of UAVs approaches a constant value at equilibrium. In the event of UAV failure, the remaining UAVs reconfigure to a new fixed formation and maintain collision-free tracking of the target's velocity, demonstrating the robustness of our control approach to failure. We validate this control approach on different simulated scenarios in MATLAB and the Gazebo robot simulator. We also experimentally test the performance of the control approach on physical robots, using Crazyflie quadrotors as the UAVs and a Turtlebot3 Burger robot as the moving ground target. The simulation and experimental results demonstrate the effectiveness of our control approach at collision-free tracking and its robustness to UAV failure.

ABSTRACT. This work proposes the use of a collection of universal global inputs (UGIs) that can be applied to a swarm of agents with only mild sensing and communication requirements to induce various emergent behaviors. The agents are not assumed to have any global positioning beyond orientation and they require no outgoing communication. This ensures that the agents are inexpensive and there is no chance of overwhelming the available bandwidth as additional agents are added. Using just simple UGIs, various useful behaviors are demonstrated through simulation. A physical agent is also built to demonstrate the attainable nature of the agent's requirements.

ABSTRACT. Reflex movements of animals are so quickly that they are difficult to achieve in response to control inputs. Therefore, we propose a tensegrity structure with bistability to realize such spontaneous motions with low energy and without a complicated control system. The tensegrity spine was chosen as the basic structure because it mimics the spine of an organism and because it can be connected to a series of homogeneous segments. By adjusting the angle of the rigid member of a basic tensegrity spine, removing one pair of soft members, and moving the mounting position of one end of another soft member, we can achieve a bistable-tensegrity joint. By fitting two laser-cut acrylic plates together, we design a lighter rigid member and make soft members out of two types of silicone rubber to realize dynamic movement. We also designed a pin for attaching soft members to hard members. Finaly, we demonstrate that development robotic joint realize the specifications, study suggests that monostability and bistability of this structure, which can be freely adjusted.

ABSTRACT. Estimating the motion state of surrounding objects is one of the key technologies for object recognition in automated driving. On-board ranging sensors are able to observe surrounding objects and object states are estimated by applying time-series processing. In order to recognize both static and dynamic objects around the vehicle, the Dynamic Occupancy Grid Map (DOGM) is one of the methods to estimate simultaneously the existence of static objects and the motion state of dynamic objects using a particle filter. This paper aims to extend the DOGM by integrating the observation information by several range sensors and the digital map in order to stably estimate surrounding dynamic objects. The evaluation results show that the introduction of the proposed method improves the variance of particles during motion estimation of dynamic objects, and achieves robust motion estimation.

ABSTRACT. In this study, we propose a soft creeping robot that utilizes twisting motion to lo-comote. The proposed robot has a cylindrical duplex structure made of silicone rubber hose and a flexible shaft. A servomotor is attached to each side of the ro-bot, and the robot shrinks and expands by rotating the motors. By rotating one motor, the shape of the robot makes a spiral wave and its length shrinks. Then by rotating the other motor to loosen the spiral, the shape of the robot returns to the straight, and the center of gravity moves. By simply repeating this mo-tion, creeping behavior is realized, and the robot moves toward one direction. Al-so, as the structure of the robot is symmetric, by reversing the timing to move the motors, the robot moves toward the opposite direction. We developed a prototype to demonstrate the effectiveness of the proposed mechanism, and experiments were conducted. As a result, we confirmed that desired creeping behavior was re-alized, and the robot could move toward desired direction in one dimensional space.

ABSTRACT. Social insects, such as ants, use a wide range of pheromones as social information. In addition, they also use the presence of other ants for decision-making as the private information. In this study, we attempted to address the problem of individual decision-making by focusing on the complementary use of pheromones and the presence of other ants. Ants were allowed to join the ant trails at right angles to the trail. By comparing with the experimental situations where no ant trails/pheromones on the path, it was found that ants entered the trail tended to flow back into the trail. These results indicate that the direction of movement of joined ants on the ant trails may be determined based on contacting with other ants.

ABSTRACT. This study aimed to assess the differences in walking trajectories between similar species of Armadillidiidae. Two species of Armadillidiidae exist in Japan; Armadillidium nasatum and Armadillidium vulgare. The present study captured the walking trajectories of these two species on a two-dimensional space using an animal tracking machine known as ANTAM. The datasets were compared via statistical methods to quantify the walking trajectories of the two species. The moving speeds and diffusion based on the mean squared displacement (MSD) were assessed to characterize the two species. The moving speed was found to indicate statistically significant differences between the two species.

ABSTRACT. Self-propelled agents can interact in many different ways, including by perturbing a shared physical environment, e.g, hydrodynamic interactions. We study motile particles that are embedded into a smectic liquid crystal, locally distorting the smectic layer spacing. This results in interactions mediated by a smectic liquid crystal distortion field and corresponds to a form of "active smectic" liquid crystal. We identify several dynamical phases that emerge in different regimes of the smectic stiffness and particle reorientation time. We characterise these as (i) ballistic motion, (ii) clustering and (iii) collective motion where orientational order emerges even thought the system lacks explicit co-alignment. We further identify an order-to-disorder transition on the addition of angular noise.

ABSTRACT. Group living animals often form dominance hierarchies. A dominance hierarchy is a social structure that is described by an asymmetrical pattern of outcomes in agonistic conflicts between group members. The rank of members in a dominance hierarchy has been studied with a focus on reproductive success. However, it remains unclear which member(s) is crucial to maintain a hierarchical structure in a fission-fusion society. One hypothesis is that the higher-ranked member plays a crucial role in maintaining the social structure by interacting with lower-ranked members. This hypothesis has been argued based on the observational study in various mammals and birds. However, not so many studies have been carried experimentally. This study aimed to test this hypothesis by the temporal removal of the top-ranked individual from a captive group of large-billed crows (Corvus macrorhynchos). From group-housed crows whose ranks of which were determined prior to the experiment, we temporarily removed the top-ranked individual and compared behaviors of remaining individuals during the removal with those before the removal. We found that the number of sequential calls of the second ranked individual, assumed as a status vocal signal, increased during the removal. No such increase of calls was found during the removal of a middle-ranked individual. This result supports the hypothesis that the top-ranked individual accounts for the maintenance of the dominance hierarchy.

ABSTRACT. This paper proposes a method of inducing sheep herding with efficient and robustness by multiple sheep dogs. Recently, much attention has been paid to control methods modeled on a sheep herd by a very small number of shepherds. Usually, these control systems are connected to the Internet in some way and the system must be capable of dealing with requirements such as robot failure or hacking by malicious entities. However, except for a few studies, multiple Shepherd agents to work together efficiently have been rarely discussed. Experimental results show that the introducing repulsion gain K_(f_4 ) between a couple of two shepherds can achieve the task more quickly and reliably than the conventional method of single shepherd by increasing the number of units.

ABSTRACT. Honey bees (Apis mellifera L.) aggregate around the queen by collectively organizing a communication network to propagate volatile pheromone signals. Our previous study shows that individual bees ``scent'' to emit pheromones and fan their wings to direct the signal flow, creating an efficient search and aggregation process. In this work, we introduce environmental stressors in the form of physical obstacles that partially block pheromone signals and prevent a wide open path to the queen. We employ machine learning methods to extract data from the experimental recordings, and show that in the presence of an obstacle that blocks most of the path to the queen, the bees need more time but can still effectively employ the collective scenting strategy to overcome the obstacle and aggregate around the queen. Further, we increase the complexity of the environment by presenting the bees with a maze to navigate to the queen. The bees require more time and exploration to form a more populated communication network. Overall, we show that given volatile pheromone signals and only local communication, the bees can collectively solve the swarming process in a complex unstructured environment with physical obstacles.

ABSTRACT. In groups of whirligig beetles Dineutus discolor (Coleoptera:Gyrinidae) swimming freely on the surface of water, we observe (i) a swim speed dependent on local density v∼ρ−ν with ν≈0.4 over two orders of magnitude in density (ii) inertial delay between individual beetle's body axis and velocity direction (iii) the coexistence of MIPS-like high an low density phases corresponding to low and high self propulsion speed. By using a standard active brownian particle model (ABP) we add a local density dependent torque term, which tends to re-orientate particles towards the group’s geometric center, that functions in open space which we call corralled active brownian particles (CABPs). The fitted model then successfully recovers a MIPS-like condensed phase for N= 200 and the absence of such a phase for smaller group sizes N= 50,100

ABSTRACT. We study a bottom-up model for swarming in discrete time. Moving agents re-orientate themselves in each timestep so as to maximise the entropy associated with visual states accessible to them in the immediate future. The relative positions and orientations of all agents in the future, and this entropy, is therefore contingent on their current reorientation move of each agent, selected accordingly. We refer to this mechanism as Future State Maximisation (FSM) and argue that it should confer evolutionary fitness for a number of reasons that, in more general settings, would include resource acquisition and risk avoidance. Here we pose FSM dynamics in terms of a genuine configurational entropy, here that of the sensor array analagous to the retina. These sensors can register 1 if one or more other unit-radius agents lie in their angular field of view, otherwise 0. The entropy is calculated from the distribution of visual states over all accessible future re-orientation sequences, explicitly enumerated. The dynamics depends primarily on an integer parameter tau, the future time horizon for the states entering the configurational entropy. Simulations generated from this model at high values of tau>5 gives rise to co-aligned, cohesive swarms even though no other interactions of any kind are present, i.e. co-alignment, attraction or repulsion. The swarms are characterised by high polar order. We further investigate the role of visual occlusion in which the entropy is computed only over the (visual projection of) those agents that are visible to another agent in the present timestep. We find no qualitative changes in the behaviour when occlusion is incorporated. For smaller values of $\tau$ we observe swarm fragmentation. We use a clustering algorithm to measure the rate of fragmentation and the co-alignment in the ordered sub-clusters. We discuss possible extensions of our work from 2D to 3D.

ABSTRACT. Animal groups including human crowds exhibit a variety of self-organizing behaviors. They often show intriguing global pattern formations that spread throughout a group over a range of inter-individual interactions. Anticipation has been considered to play a key role in a wide range of self-organizing systems. However, studies on whether individual anticipation offers functional benefits to the group are surprisingly scarce. Here we showed the first link between emergent patterns of organization and individual anticipation in human crowds by experimentally manipulating pedestrians’ anticipatory abilities. We found that the manipulation (a visual distraction on some of the pedestrians) diminished global pattern formation in a crowd. Moreover, both the distracted pedestrians and the non-distracted ones had difficulties navigating. These results imply avoidance maneuvers are normally a cooperative process and that mutual anticipation facilitates the efficient transition to emergent patterning. We expect that our findings will inform future models of self-organizing biological and robotic systems where a new approach to navigation and collision avoidance is needed.

ABSTRACT. In a seed-harvester ant colony, an appropriate level of collective alarm response was needed to mitigate signal propagation in the nest when we initiated a non-colony-threatening event by re-introducing 3 agitated ants instead of real threats. An appropriate collective alarm response for a social group can be evaluated in two following aspects: rapidly responding to perturbations and damping alarm response after discerning non-colony-threatening perturbations. Here, we developed an agent-based discrete-time Markov chain model to simulate alarm signal propagation and dampening in a subgroup of worker ants and used this model to investigate how individual alarm response threshold and persistence of alarm behavior modulate the collective response.

ABSTRACT. This paper discusses the problem of how to generate a stable limit cycle gait of a point-footed walker while the stance foot slides on a low friction road surface. First, we introduce a 6-DOF walking robot model and its equations of motion and collision. Second, we design a control system and propose an algorithm for calculating the target upper-body angle and steady initial state. Furthermore, we propose another algorithm for calculating the target amplitude of vibration of the upper body so that the amount of change in the stance foot for one step becomes zero. The validity of the proposed methods is investigated through numerical simulations.

ABSTRACT. This paper discusses the condition necessary for achieving instantaneous stance-leg exchange of a 3-link point-footed biped that walks on a low-friction road surface. First, we introduce a 5-DOF biped robot model and develop the equations of motion and collision. Second, we mathematically and numerically identify the parameter conditions that the rear foot leaves the ground immediately after impact by using an analytical solution of the impulse. Third, we design a control system for generating a stable sliding limit cycle gait by setting the relative hip-joint angle and upper-body angle as the control outputs. Furthermore, we conduct numerical simulations to investigate the fundamental properties of the typical sliding limit cycle gait.

ABSTRACT. In this paper, we introduce an extremely simple and efficient control method for robot dynamic walking. To conveniently control the walking pace, the walker is entrained to a periodic input torque with the frequency near its natural oscillation. A comparison among different waveforms is numerically conducted via the Arnold tongue. The tendency of the entrainability of the input waveform affected by the shape of it is observed accordingly.

ABSTRACT. Biped robots have been attracted great attention for exploring the mechanism of human walking and are expected to provide a wide range of application scenarios. Nevertheless, the stable and efficient gait generation as the first step is not easy to implement. In an effort to overcome this problem, an indirect method has been recently introduced to leverage the natural dynamics of the passive dynamic walking. In this work, we introduce a more straightforward approach to strengthen the implementability of it. First, we introduce the typical modeling and control method of the compass-like robot with semicircular feet. Second, we conduct the numerical simulations to observe the typical gait, and analyze the results of it. Moreover, we investigated the effect of feedforward control on travel performance.

ABSTRACT. In this paper, we conducting the experiment of loupe-shaped object transporting system by means of shaft rotational friction. Based on the previous work, we have acquiring the function which is suitable to express the conveying speed of the object. Meanwhile, we utilizing fitting of a polynomial to indicate the relations among conveying speed，shaft rotational speed and tilt angle.

ABSTRACT. This paper investigates the sliding motion generation for an underactuated locomotion robot through analysis of relation between friction force and motion of the center of mass. The robot consists of a body frame, an inclined upper link having a telescopic joint and a weight driving by the telescopic joint. First, we introduce the mathematical model of the robot. Second, we mathematically analyze the condition of sliding generation and of the body rotation. Third, we show results of numerical simulation with respect to the inclination angle of the telescopic joint. We confirmed that the generated motion accords with the results of mathematical analysis.

ABSTRACT. To address the problem of stable walking of underactuated biped robots, an adaptive feedforward control strategy is proposed. First, the bipedal robot with the circular foot is modeled and its equation of motion and collision are derived. Second, a feedforward control strategy based on the entrainment effect is proposed, which realizes the control of the robot's own walking cycle by controlling the change of the input cycle of the given track from the perspective of natural dynamics, and then achieves stable walking. Furthermore, to explore the accuracy of the simulation results, we built an experimental machine and conducted preliminary experiments.

ABSTRACT. Passive dynamic walking is a model that walks down a shallow slope without any control or input. This model has been widely used to investigate how stable walking is generated from a dynamic viewpoint, which is useful to provide design principles for developing energy-efficient biped robots. The basin of attraction shows fractal even for a single attractor. The passive dynamic walking shows chaotic attractor through a period-doubling cascade by increasing the slope angle and the chaotic attractor suddenly disappears at a critical slope angle. In our previous work, we investigated the disappearance in the viewpoint of the basin of attraction and found that the stretch-bending deformation produces a Cantor set. In connection with this finding, it has been suggested that the chaotic attractor disappears by the boundary (attractor) crisis, which is a general phenomenon in chaotic dynamical systems occurred by homoclinic and heteroclinic tangencies. However, the detailed mechanism in the passive dynamic walking remains unclear. In this study, we computed stable and unstable manifolds of saddle points of the Poincaré map and clarified that the boundary crisis occurs by a heteroclinic tangency based on the formation mechanism for the basin of attraction.

ABSTRACT. In human locomotion, walk and run coexist in intermediate speeds, which may be considered as the coexistence of limit cycles of walk and run from dynamical systems point of view. A simple model has indicated a region of solutions leading the system to fall that separates the basins of the limit cycles. This implies the absence of a branch of unstable periodic orbit that connects the branches of the attractors in the bifurcation diagram, unlike the usual view in dynamical systems theory. To do mathematical analysis of this nontrivial coexistence, we study a yet simpler vertical up-and-down motion of a spring-mass model. The hybrid dynamical system model shows not only the coexistence of two limit cycles but also the basin of fall. The mechanism of the coexistence and the bifurcation are analyzed by considering the phase space structure.

ABSTRACT. This paper presents a novel gait of a snake robot without wheels to move on a pipe with branches. The robot touches the pipe with some parts of its body and lifts up other parts of its body to make it ungrounded. By using shift motion and rolling motion, the contact point between the robot and the pipe move freely, and the robot can move fast on a pipe with branches. The effectiveness of the proposed gait is demonstrated by simulations.

ABSTRACT. This paper proposes a novel control method for a snake robot that is robust to modeling errors. The proposed method applies the port-controlled Hamiltonian based approach for path-following control, which has originally been proposed for a full-actuated friction-less system. This paper extends the controller for an under-actuated system with friction. The validity of the controller is checked through simulations.

ABSTRACT. Although significant progress has been made in the development of snake-like robots, their motion control, gait generation and adaptation to environmental constraints still presents a major research challenge. This extended abstract and the associated talk presents a body of research, previously published across three papers, culminating in the development of a snake-like robot with highly adaptive undulatory locomotion. The robot’s control system is an abstracted neural model derived from a detailed and biologically realistic neuro-mechanical model of forward locomotion in the nematode worm Caenorhabditis elegans. The neuro-mechanical model was initially used in a scientific con-text to advance our understanding of locomotion control in this popular model organism, and generated testable predictions about the roles and properties of its motor neurons. The model was informed by a wealth of information that al-ready existed about this well-characterized organism, along novel behavioural experiments by the author and his collaborators

ABSTRACT. Snake robots are expected to be used for work in places where it is difficult for humans to enter. In this study, the authors estimates parameters of a snake robot moving by helicoidal rolling motion in a pipe by using Extended Kalman Filter(EKF). The authors verified the effectiveness of the algorithm by conducting experiments.

ABSTRACT. In this research, we apply shape-based compliance control (SBC) to our snake robot that can measure the magnitude of circumferential pressure by Center of Pressure (CoP) sensor and moves by helicoidal rolling motion. Effectiveness of the snake robot is verified by conducting experiments.

ABSTRACT. This paper proposes an evolutionary robotics approach that applies echo state networks as controllers for a robotic swarm. The main characteristics of the echo state network are that the hidden layer is generated with sparse and random connections, and only the weight values that are connected to the output are trained. This paper applies an evolutionary algorithm to optimize the output weight values of the echo state network. We compared the performance of the controller using the echo state network with a traditional neural network controller. The performance of the robotic swarm is evaluated in a path-formation task using computer simulations. The results showed that the echo state network could reduce the effort required in designing the parameters of the robot controller. Besides, the controller using the echo state network showed a faster convergence in optimizing controllers than that of the traditional recurrent neural networks.

ABSTRACT. This paper applies a topology and weight evolving artificial neural network (TWEANN) to design controllers for a robotic swarm. A typical method for designing controllers using an evolutionary robotics approach applies neural networks as the robot controllers and only optimizes the weight values of the neural network. However, this approach might restrict the robot behavior or might have redundant structures within the robot controller. In this paper, we applied the mutation-based evolving artificial neural network (MBEANN), which is a TWEANN algorithm that only employs mutation to evolve neural networks, to design the controller for a robotic swarm. For comparison with the MBEANN approach, we used the NeuroEvolution of Augmenting Topologies (NEAT), which is a widely used TWEANN algorithm. The controllers are evaluated in cooperative transport performed by a robotic swarm using computer simulations.

ABSTRACT. This paper focuses on developing the collective step-climbing behavior of a multi-legged robotic swarm. Most studies on swarm robotics develop collective behavior in a flat environment using mobile robots equipped with wheels. However, mobile robots with wheels could only show relatively simple behavior, which limits a task that could be ad- dressed by a robotic swarm. In this paper, a step-climbing task is em- ployed, which aim to climb a step that is high for a single robot to climb alone. The robots have to use other robots as a foothold to climb the step. The robot controllers are obtained by the combination of the neuroevolution approach and manual designed methods. The results of the computer simulations show that the designed robots and controllers successfully achieve the step-climbing task.

ABSTRACT. This paper investigates how well human operators can control a robot, the task of which is leading a flock of mobile robots to their target area. An operator teleoperates a robot with limited sensing ability to lead as many robots as and to drive as fast as possible toward the goal area. The teleoperated robot has cameras on the front and rear as sensing devices and sends the local information perceived by either of them to the operator. Flocking robots make decisions using a Boid model and can learn to coordinate their behavior. The flocking performance is evaluated in computer simulations.

ABSTRACT. In this study, we focus on a multi-robot task of moving along a wall while achieving formation. This task involves two types of rolls assigned to robots: moving along and away from the wall. To perform the task, we design a distributed controller using the gradient method. By using the designed controller, each robot can autonomously determine its own roll, based on the distance to the wall. Besides, the designed controller requires only relative positions acquired with sensing devices such as LiDAR, but not absolute positions with communication devices. The effectiveness of the designed controller is verified through experiments with two-wheeled mobile robots.

ABSTRACT. Once a building collapses due to an earthquake, fast and thorough search for persons in need of rescue is required to save their lives. Cyborg insects, that is, insects equipped with devices for sensing, wireless communication, and movement control, are promising tools to explore a collapsed building which it is hard for rescuers to enter. In this paper, we propose algorithms to enable cyborg insects to search a collapsed building effectively and efficiently in a fully distributed and autonomous manner. Taking inspiration from flocking models, algorithms determine the direction of movement by combinations of four attractive or repulsive forces for cyborg insects to move apart from each other and toward unvisited areas. Through simulation experiments we show that an algorithm combining repulsive forces from nearby cyborg insects and the starting point is the most effective to achieve as high as 76.7\% searched rate after five hours exploration.

ABSTRACT. Car sharing has become a popular as a new form of transportation mode in cities. In particular, station-based one-way car sharing gives customers the option to return the vehicle to any available station. It has been gaining popularity owing to its convenience, particularly in Japan where it is currently being tested. However, under this type of one-way car-sharing system, user demand is not constant across stations, which leads to the uneven distribution of vehicles among stations. The cost of eliminating this uneven distribution is a major problem in terms of profitability. The goal of this research is to solve the uneven distribution problem of vehicles in a one-way car-sharing system without rebalancing (managed redistribution of vehicles), taking into account realistic constraints on the number of vehicles. The main strategy is real-time pricing, with which the user demand moves to cheaper links. This paper provides a model of this system to evaluate the effectiveness of the real-time pricing.

ABSTRACT. This paper discusses coverage controllers for multi-agent systems inspired by the chemotaxis of microorganisms that the author proposed in a previous study. In particular, we focus on the performance of the controllers quantified as an achieved degree of coverage and conduct its theoretical analysis.

ABSTRACT. Monitoring of crop fields to map features like weeds can be efficiently performed with unmanned aerial vehicles (UAVs) that, owing to their privileged perspective and motion speed, can cover large areas in a short time. However, the need for high-resolution images for precise classification of features (e.g., detecting even the smallest weeds in the field) contrasts with the limited payload and flight time that characterizes current UAVs, and requires several flights to uniformly cover a large field. However, the assumption that the whole field must be observed with the same precision is unnecessary when features are heterogeneously distributed, like weeds appearing in patches over the field. In this case, an adaptive approach that focuses only on relevant areas can perform better, especially when multiple UAVs are employed at the same time. Leveraging on a swarm-robotics approach, we propose a monitoring and mapping strategy that adaptively chooses the target areas based on the expected information gain, which measures the potential for uncertainty reduction due to further observations. The proposed strategy scales well with group size and overall leads to smaller mapping errors than optimal pre-planned monitoring approaches.

ABSTRACT. We study the collective motion of autonomous mobile agents on a ringlike environment. The agents' dynamics is inspired by known laboratory experiments on the dynamics of locust swarms. In these experiments, locusts placed at arbitrary locations and initial orientations on a ring-shaped arena are observed to eventually all march in the same direction. In this work we ask whether, and how fast, a similar phenomenon occurs in a stochastic swarm of simple agents whose goal is to maintain the same direction of motion for as long as possible. The agents are randomly initiated as marching either clockwise or counterclockwise on a wide ring-shaped region, which we model as $k$ ``narrow'' concentric tracks on a cylinder. Collisions cause agents to change their direction of motion. To avoid this, agents may decide to switch tracks so as to merge with platoons of agents marching in their direction.

We prove that such agents must eventually converge to a local consensus about their direction of motion, meaning that all agents on each narrow track must eventually march in the same direction. We give asymptotic bounds for the expected amount of time it takes for such convergence or ``stabilization'' to occur, which depends on the number of agents, the length of the tracks, and the number of tracks. We show that when agents also have a small probability of ``erratic'', random track-jumping behaviour, a global consensus on the direction of motion across all tracks will eventually be reached. Finally, we verify our theoretical findings in numerical simulations.

ABSTRACT. We present a novel approach for informative path planning using multi-robot systems in the context of mapping tasks under time and communication constraints. The approach is based on modeling the task as a regression problem using Gaussian processes (GP) and adaptively directing the robots towards the regions that are most informative for GP learning. The methodology is based on a multi-stage process where either a robot or a group of robots search for the best convex region where to sample new data, identify the most informative sampling locations in the region, and compute an optimized path through them. The process is iterated over time adapting to newly gathered evidence and is performed in a collaborative manner among the robots. Overall, techniques from Monte Carlo, sequential Bayesian inference, and orienteering optimization models are combined in an integrated strategy. Fully distributed and leader-follower architectures are designed to implement the multi-stage strategy, and have been evaluated in simulation studies, showing up to 69% of improvement over a baseline strategy.

ABSTRACT. Animal formation movements, from bird flocks to fish schools, are enabled by intriguing mechnimsms that utilize local sensing signals to realize global collective motion coordination. The design of formation control strategies for teams of mobile robots can benifit from the better understanding of how animals implement sensing or communication topologies within groups. Along this line of research, rigidity graph theory turns out to be a powerful tool to gain insight into how multi-agent structures become rigid under inter-agent distance or angle constraints. In this talk, I will report recent developments on how such rigidity properties of multi-agent formations may lead to different local and global behaviors when the available types of inter-agent sensing signals change. In particular, I emphasize the importance of formation robustness against sensing noises and correspondingly the effectiveness of estimator-based formation control.

ABSTRACT. In the animal evolution, functions of multicellular individuals have been adaptive, organizing elaborate systems to efficiently survive and reproduce. Among diverse animal lineages, some animals have acquired social/colonial systems in which individuals can differentiate into multiple types with specific morphologies and functions. Eusocial insects and colonial marine animals are representative animals that represent such systems. I have been studying the developmental and physiological mechanisms underlying caste differentiation, especially in termites. In termite societies, multiple types of functional individuals, that is, castes, perform divisions of labors to coordinate social behaviors. Among other castes, the soldier caste is distinctive since it is sterile and exclusively specialized into defensive behavior with largely modified morphological features. In this talk, I will summarize our previous studies focusing on the social systems in termites, including caste differentiation, reproductive division of labor and behavioral regulations. Furthermore, I would like to report recent advances on social systems in marine colonial animals like bryozoans. We are focusing on the developmental and physiological systems of colonial systems in some bryozoan species which produce distinctive phenotypes specialized in defensive tasks like termite soldiers. In this talk, some interesting behavioral and/or reproductive phenomena seen in marine animals will also be introduced.