RM18: SWEDISH CONTROL CONFERENCE 2018 (REGLERMöTE 2018)
PROGRAM FOR TUESDAY, JUNE 19TH
Days:
next day
all days

View: session overviewtalk overview

09:15-10:00 Session 1: Plenary 1: 5G networks – the new communication fabric

Magnus Frodigh

Head of Ericsson Research

Abstract:The coming 5G networks will be a platform for a large number of communication applications. Higher data rates, shorter latencies, quality of service and security will accommodate many more demands than just voice communication and internet access. With a cloud moving towards the edge, compute and storage resources will move closer to the end user – or maybe we should say “end machine”, as we will se 5G coming into play in many industrial applications where for instance robot control loops can take place wirelessly over 5G

Location: Q1
10:20-12:00 Session 2A: Aerial vehicles
Location: Q1
10:20
Autonomous cooperative flight of rigidly attached quadcopters

ABSTRACT. Research within the area of Unmanned Aerial Vehicles (UAVs) is a continuously expanding field due to their wide and diverse range of applications such as inspection, monitoring, mapping, precision agriculture, aerial imaging and entertainment. However, their application in tasks such as transportation, tool manipulation or assistance in emergency situations still poses many challenges.

In this extended abstract, a novel strategy for cooperative flight of physically attached quadcopters which allows the quadcopters to perform the transportation of heavier loads and carrying out more complex tasks is proposed. Specifically, a method for online and automatic modelling and control of the quadcopters without human intervention is introduced.

First, the method performs the estimation of the physical structure attaching the quadcopters' solely relying on information from the quadcopters Inertial Measurement Unit (IMU) obtained via simple and short online experiments. Then, given the estimated physical structure, a stable operation of the quadcopters is achieved via a distributed controller, where the controller parameters are obtained via reinforcement learning. Finally, experimental results validate the proposed method, showing that a correct estimation of the physical structure is obtained as well as a stable flight is achieved for a set of connected quadcopters.

10:40
Cooperative UAVs as a tool for aerial inspection of the aging infrastructure

ABSTRACT. This article presents an aerial tool towards the autonomous cooperative coverage and inspection of a 3D infrastructure using multiple Unmanned Aerial Vehicles (UAVs). In the presented approach the UAVs are relying only on their onboard computer and sensory system, deployed for inspection of the 3D structure. In this application each agent covers a different part of the scene autonomously, while avoiding collisions. The visual information collected from the aerial team is collaboratively processed to create the 3D model. The performance of the overall setup has been experimentally evaluated in a realistic outdoor infrastructure inspection experiments, providing sparse and dense 3D reconstruction of the inspected structures.

11:00
Verification of cooperative maneuvers in flightgear

ABSTRACT. We develop a simulation setup for testing and analyzing cooperative maneuvers and corresponding control algorithms. We also find feasible initial sets using backwards reachable set computations for the cooperative control problem, which we then test using the simulation setup. The particular example considered is a cooperative rendezvous between a fixed-wing unmanned aerial vehicle and a unmanned ground vehicle. The open-source software FlightGear and JSBSim are used for the vehicle dynamics, enabling testing of algorithms in a realistic environment. The aircraft models include nonlinear, state-dependent dynamics, making it possible to capture complex behaviors like stall and spin. Moreover, environmental effects such as wind gusts and turbulence are directly integrated into the simulations. From the simulations we get a comprehensive understanding of the controller performance and feasibility when tested in a real- time scenario. Results from several landing simulations are presented, and demonstrate that the MPC solution for the cooperative rendezvous problem is a promising method also for use in complex, safety-critical systems.

11:20
Bar stabilization tethered to two non-identical aerial vehicles

ABSTRACT. We consider a system composed of a bar tethered to two unmanned aerial vehicles (UAVs), where the cables behave as rigid links under tensile forces, and with the control objective of stabilizing the bar's pose around a desired pose. Each UAV is equipped with a PID control law, and we verify that the bar's motion is decomposable into three decoupled motions, namely a longitudinal, a lateral and a vertical. We then provide relations between the UAVs' gains, which, if satisfied, allows us to decompose each of those motions into two cascaded motions; the latter relations between the UAVs' gains are found so as to counteract the system asymmetries, such as the different cable lengths and the different UAVs' weights. Finally, we provide conditions, based on the system's physical parameters, that describe good and bad types of asymmetries. We present experiments that demonstrate the stabilization of the bar's pose.

11:40
Guidance, navigation and control design for the e.Deorbit space debris removal mission

ABSTRACT. The e.Deorbit mission is an ESA active debris removal mission aimed at deorbiting the large inoperative satellite Envisat. A spacecraft will capture Envisat with either a net or a robotic arm and will, after a successful capture, deorbit itself and Envisat. The capture of the non-cooperative target poses many new challenges on the Guidance, Navigation and Control (GNC) system. This extended abstract gives an overview of the challenges faced and the proposed GNC design with its mission phases, controller and filter design. A fuel minimizing Model Predictive Controller is developed to control the relative position between the spacecraft and the target while guaranteeing a collision free capture. The MPC controller is highly versatile and can be used in all mission phases. The proposed GNC design is able to successfully perform all critical mission phases.

10:20-12:00 Session 2B: Human applications
Location: Q2
10:20
Control-engineering perspective on deep brain stimulation: Revisited

ABSTRACT. Deep brain stimulation (DBS) is a an established therapy in neurological and mental disorders making use of electrical pulses chronically delivered to a certain neural target through surgically implanted electrodes. Different stimulation targets are selected for treating different diseases. The therapeutical effect of DBS is highly individual and depends on the target coverage by the stimuli and the amount of spill beyond it. This can be suitably formulated as an optimization problem. The train of DBS pulses has three tunable degrees of freedom, namely the amplitude, frequency, and pulse width (duty cycle). Naturally, the DBS neither cures the disease nor stops its progression but only alleviates the symptoms. Since the biological mechanism underlying the DBS therapy is mainly unknown, and due to high inter-patient and intrapatient variability of the DBS effect, a pragmatic approach to the DBS programming is to see it as a control system for the symptoms. Such a technology assumes that the symptoms are accurately quantified. This paper is a followup of a presentation at ACC 2016, where a control-engineering view of DBS has been introduced. Now, two years later, more insights into the therapy itself as well as into the mathematical models behind its optimization and individualization have been obtained. Further, the proposed DBS programming approach has been validated in a limited clinical study. This publication summarizes the progress in the individualized DBS and sets up new goals for the development of this steadily improving and expanding therapy.

10:40
Identification of planar human movement using kinematic constraints and inertial sensors

ABSTRACT. Planar inverted pendulum (PIP) models are often used to describe the dynamics of human standing. However, it is only a good approximation if the human is standing still and the movement is mostly constrained to one plane. Using the kinematic constraints of such PIP models to formulate a cost function, and measurements from inertial measurement units (IMU) we formulate an optimization problem to estimate the plane of movement. Furthermore, the residuals of the cost function can be used to classify movement which conform to a PIP model. We demonstrate the proposed method's functionality using real data and discuss it's potential use and future work.

11:00
Data-driven modelling of pelvic floor muscles dynamics

ABSTRACT. This paper proposes individualized, dynamical and data-driven models that describe pelvic floor muscle responses in women that use vaginal dilation. Specifically, the models describe how the volume of an inflatable balloon inserted at the vaginal introitus dynamically affects the aggregated pressure exerted by the pelvic floor muscles of the person. The paper inspects the approximation capabilities of different model structures, such as Hammerstein-Wiener and NARX, for this specific application, and finds the specific model structures and orders that best describe the recorded measurement data. Hence, although the current dataset is drawn from a sample of healthy volunteers, this paper is an initial step towards better understanding women’s responses to vaginal dilation and facilitating individualised medical vaginal dilation treatment.

11:20
Reducing the human effort for human-robot cooperative object manipulation via control design

ABSTRACT. This study is concerned with the shared object manipulation problem in a physical Human-Robot Interaction (pHRI) setting. In such setups, the operator manipulates the object with the help of a robot. In this paper, the operator is assigned with the lead role, and the robot is passively following the forces/torques exerted by the operator. We propose a controller that is free from the well-known translation/rotation problem and enhances the operator's ability to move the object by reducing the human effort. The key point in our study is that the controller is defined based on the instantaneous center of rotation. The passivity of the system including the object and the manipulator has been evaluated. Simulation results validate the theoretical findings on different scenarios of subsequent rotations and translations of the object.

11:40
Multi-sensor hearing aid research

ABSTRACT. Miniaturization of electrical devices are paving the way for more sensors in future hearing aids. This can for example enable automatic source separation and tracking, hearing threshold measurements, and eye gaze tracking for hearing aid control. This paper presents some promising sensors and sensor combinations which are, or can be, used for hearing aid research.

10:20-12:00 Session 2C: Biology applications
Location: Q31
10:20
Identification of rational regression models

ABSTRACT. Models rational in the parameters arise frequently in applications. As with all models that are non-linear in the parameters, direct parameter estimation, using e.g. non-linear least squares, can then become challenging due to the issues of local minima and finding good initial estimates. Here we propose a least-squares-based multi-step method which we conjecture will give consistent estimates. The proposed method is illustrated by an extended Monod-type model. Numerical simulations are conducted to demonstrate the advantages of the proposed method.

10:40
Comparing water treatment topologies in recirculating aquaculture plants

ABSTRACT. Aquaculture, the farming of fish and aquatic crops such as kelp and algae, is traditionally carried out in natural bod- ies of water. An alternative is land-based farming in tanks or raceways, which is particularly attractive when coupled with water treatment to form a recirculating aquaculture system (RAS). Benefits compared to traditional farming in open cages include reduced emissions of nutrients, small or no risk of escapes, and control of pathogens (Thorarensen and Farrell, 2011). Water treatment takes place in a series of mechanical filters and biological reactors, where particulate and dis- solved matter is degraded by microorganisms similarly to how municipal sewage is treated. The biological nature of recirculating aquaculture systems makes experimental process development troublesome. Contributing factors in- clude very long time constants, biological variations, and concerns for animal welfare. This strongly motivates the use of dynamic simulations, and for that purpose a RAS simulator – called FishSim – was developed (Wik et al., 2009). However, the capabilities of that implementation were limited by numerical problems. Using Modelica, a high-level object-oriented language for dynamic systems modeling (Modelica Association, 2012), we have developed a new simulation tool for recirculating aquaculture. Like FishSim, it is based on Activated Sludge Model 1 (Henze et al., 2000), but this implementation is numerically well-behaved and robust which allows a much greater variety in the simulated systems. It is also significantly faster, even after the models have been ex- panded with many more features, such as energy balances, different feeding options, and a separation of autotrophic bacteria into ammonia-oxidizing and nitrite-oxidizing bac- teria. Since open-source Modelica tools are available, the software is also free to use. Water treatment is central in recirculating aquaculture. Fish excrete ammonia, which is toxic to them. Aerated bioreactors are typically employed to remove ammonia and ammonium via nitrifying autotrophs, which require low levels of biodegradable organics to thrive. Nitrifica- tion creates nitrite (also toxic to fish at low levels) and nitrate, the latter which is removed by water exchange or denitrification. Denitrifcation conversely requires high availability of biodegradable organics, but only progresses rapidly in the absence of oxygen. The treatment systems often further contain particle filters and UV and/or ozone treatment against pathogens. While it is reasonably clear to the industry which compo- nents should be present in the treatment system, the order in which they are best employed is still an open question. In the literature and supplier information material there is a large number of suggested configurations, but few studies comparing them. Some guesses can be made based on elementary chemical reaction engineering, but the very complex dynamics of the biological treatment leads to high uncertainty. Using the simulator, we have investigated and compared several treatment topologies. Through parameter opti- mization based on a genetic algorithm (Haupt and Haupt, 2003) the minimal reactor sizes in each configuration was found which could maintain acceptable levels of ammonia and nitrate. The resulting sizes are an indicator of which topology is the most effective.

11:00
Poly-pathway dynamic models of the production process of therapeutic drug protein

ABSTRACT. The pharmaceutical industry is the most profitable industry. On a world-basis, this sector has the highest level of R&D intensity, 2-3 times higher compared to the average of all industry. This last decade biopharmaceuticals (i.e. therapeutic drug protein) have taken leading positions worldwide for revenues, in particular thanks to therapeutic antibodies. These latter are produced by mammalian cells, which are highly complex. Models could significantly decrease the R&D costs and be the basis of manufacturing process control. We aim at developing dynamical models of biopharmaceutical manufacturing cultivation process with the purpose of optimizing the feed strategies for batch or continuous processes, and ultimately achieving model-based control. These models are based on the detailed description of reactions taking place in the cells accompanied by the kinetics of these reactions. The measured variables are the concentrations of the components of the culture (i.e. extracellular components), while the information of the intracellular components is not available in dynamic systems. To create dynamical models based on known variables, elementary flux mode can be used. However these models are currently limited in the number of reactions, the complexity of the kinetics and the identification of the model parameters. We have created a new approach for the field of biotechnology using column generation, originally developed for vehicle and aircraft routing. With this method, the number of reactions is not limited, and instead a reduced model is created including the reactions relevant for the experimental data set. The model identification is based on using data generated in steady state where only a defined number of stimuli is applied generating a response directly linked to this stimuli to the contrary of the approach normally used of batch culture where many parameters are varying simultaneously. The obtained poly-pathway model includes the different pathways that the cells can use, including kinetics of activation and inhibition, and providing the simulation of the metabolism in mammalian cell process. We will present this modelling approach illustrated by case study.

11:20
Light spectrum optimization for plant growth using biological feedback

ABSTRACT. The use of light emitting diods (LEDs) as greenhouse illumination is increasingly common. When each LED color is individually dimmable both light spectrum and light intensity can be tuned, which opens up for optimisation of photosynthesis through automatic control of light quality and quantity. However, this requires a non-destructive biological growth signal that can be measured fast, remotely and preferably without interacting with the plants. A potential candidate signal is steady-state chlorophyll a fluorescence gain at 740 nm, defined as dF740/dq, i.e. the difference in fluorescence at 740 nm divided by the difference in incident light quanta caused by a (small) change in intensity of each individual LED color in the lamp (Ahlman et al., 2017). By automatically adjusting the spectrum, to aim for equal fluorescence gains for all LED colors (Wik et al., 2014), the instant photosynthetic rate can be optimised given a preset electric power input to the lamp. When implementing such a controller though, constraints on the spectral distribution are needed to minimise a negative impact on plant morphology.

In this study measurements were conducted (on cucumber and lettuce) under different background light, and at each setting excitation signals were sequentially added by each of six different LED colors (peak wavelength at 400, 420, 450, 530, 630 and 660 nm). The corresponding changes in steady-state fluorescence were measured with a spectrometer and the fluorescence gain (dF740/dq) was calculated for each LED color and at each background light setting. These fluorescence gains were compared in order to evaluate the different LEDs' relative effect on photosynthesis under each of the different background light settings. Additionally, the photosynthetic rate was measured by an infrared gas analyzer (IRGA) in a few sets and (in the same manner) the photosynthetic rate gain was calculated. The light regimes investigated ranged from 160 to 1000 mumolm-2s-1. At each intensity level five different spectral distributions were applied, composed of different amount of red light, ranging from 50 to 100% of the total photon flux density and with a sustained green to blue ratio (g:b 1:2).

The mutual relation between the fluorescence gains corresponding to the different LED colors did not change significantly due to background light quality. The fluorescence gain was found to be highest in response to red LEDs (660 followed by 630) for both cucumber and lettuce and for all light settings. However, the relative size of the gains did change due to light quantity. Furthermore, it was found to be species dependent. These observations, i.e. the relative efficiency of enhancing the photosynthesis of the different LED colors, are in agreement with the action spectrum for cucumber and lettuce found by McCree, 1972.

An online controller, working continuously to find the optimal light spectrum given a preset electric power input to the lamp, seems not to be needed. However, we expect regular monitoring of the fluorescence gain throughout the growth cycle, to be useful. For example, to detect diverse degradation of the LEDs, or to identify where the light curve saturates. In previous work (A.-M. Carstensen et al., 2016) we have shown that the dynamics of the fluorescence response to changed incident light, varies with the status of the plant, for example due to light inhibition. Further research aima at identifying if light inhibition, salt stress, biotic stress and draught, can be observed as changes in the mutual relation of the steady-state fluorescence gains.

References
Ahlman, L., D. Bånkestad, and T. Wik (2017). Using chlorophyll a fluorescence gains to optimize LED light spectrum for short term photosynthesis. Computers and Electronics in Agriculture 142. Part A, pp. 224-234.
Carstensen, A.-M. et al. (2016). Remote detection of light tolerance in Basil through frequency and transient analysis of light induced fluorescence. Computers and Electronics in Agriculture 127, pp. 289-301.
McCree, K. J. (1972). Action spectrum, absorptance and quantum yield of photosynthesis in crop plants. Agricultural Meteorology 9, pp. 191-216.
Wik, T., A.-M. Carstensen, and T. Pocock (2014). Spectrum optimization for artificial illumination. WO Patent App. PCT/EP2013/069,820.

11:40
Attractivity of the synchronous mode in hybrid observers for the impulsive Goodwin's oscillator subject to harmonic exogenous excitation

ABSTRACT. The paper deals with basin of attraction analysis of two observers estimating the states of a hybrid version of Goodwin's oscillator forced by a continuous exogenous signal. The continuous linear part of the observed plant is controlled by an intrinsic impulsive feedback that contributes discrete dynamics to the closed-loop system. The impacting pulsatile feedback signal is not available for measurement and, therefore, has to be reconstructed. Under harmonic exogenous excitation, the impulsive Goodwin's oscillator exhibits bistability, which phenomenon significantly complicates observer design. The attractivity of a null solution of the hybrid state error estimation dynamics (termed as synchronous mode) has to be maximized in order to cover all possible initial state estimate deviations. A detailed analysis of the synchronous mode for a previously considered observer reveals a considerable asymmetricity in the basin of attraction relative to a jump instant and susceptibility to overjumps. This may result in the observer converging a stable stationary mode distinct from a synchronous one, for some initial conditions. To circumvent this, a new hybrid observer with a more elaborate frequency modulation law in the output error feedback structure is proposed. Numerical analysis suggests that the new observer is able to approach the synchronous mode from all admissible initial estimates.

10:20-12:00 Session 2D: Nonlinear control
Location: Q34
10:20
Computation of rational parameter dependent Lyapunov functions for LPV systems

ABSTRACT. In this contribution, we present a computational method for the global stability analysis of linear parameter varying systems under rational parameter dependence. Using the linear fractional representation (LFR) of the system equation, we generate a set of rational basis functions, which will give the structure for the parameterized rational Lyapunov function. Based on the earlier results of Trofino and Dezuo (2013), affine parameter dependent linear matrix inequalities (LMIs) are formulated to ensure the Lyapunov conditions and hence asymptotic stability.

10:40
Sliding mode control for hydraulic actuated cranes.

ABSTRACT. Two new Sliding Mode algorithms are used for control hydraulics systems which are known to have highly non-linear dynamics. Two main advantages are emphasized: chattering alleviation and ease of implementation without system parameters knowledge or estimation procedures.

11:00
On stability for state-lattice trajectory tracking control

ABSTRACT. To guarantee that an autonomous vehicle is operating safely, efficient tools for verifying stability of the closed-loop system need to be developed. The key components for the two lowest levels of control in a self-driving vehicle are the controlled vehicle, the low-level controller and the local planner. The local planner that is considered in this work constructs a nominal trajectory by combining a finite number of precomputed motions. When this motion planner is considered, we show that the closed-loop system can be modeled as a nonlinear hybrid system. Based on this, we propose a novel method for analyzing the behavior of the control error, how to design the low-level controller and how to potentially impose constraints on the local planner, in order to guarantee that the control error is bounded and decays towards zero. The proposed method is applied on a truck and trailer system and the results are illustrated in simulation examples.

11:20
Restricted-complexity characterization of control-invariant domains with application to lateral vehicle dynamics control

ABSTRACT. This paper proposes two algorithms to find a restricted-complexity robust control invariant (RCI) set along with a state-feedback gain. These algorithms are applicable to a linear system with additive disturbances subject to found polytopic state and input constraints. The RCI set is a polytope with restricted complexity, symmetric around the origin. Using a state transformation, novel LMI conditions are derived for the system constraints and invariance condition. Moreover, a new approach is proposed to iteratively increase the volume of the computed RCI set. The effectiveness of the proposed algorithm is illustrated by using lateral vehicle dynamics control example.

11:40
Feedback linearization compared to Jacobian linearization for LQ control of an industrial manipulator

ABSTRACT. Feedback linearization is compared to Jacobian linearization for LQ control of a two-link industrial manipulator. A method to obtain equivalent nominal performance for both designs is introduced. Results do not show any conclusive advantages of Feedback linearization.

13:00-14:30 Session 3: Interactive session I
Location: Q33 & Q36
13:00
Minimizing long vehicles overhang exceeding the drivable surface via convex path optimization

ABSTRACT. This paper presents a novel path planning algorithm for on-road autonomous driving. The algorithm targets long and wide vehicles, in which the overhangs (i.e., the vehicle chassis extending beyond the front and rear wheelbase) can endanger other vehicles, pedestrians, or even the vehicle itself. The vehicle motion is described in a road-aligned coordinate frame. A novel method for computing the vehicle limits is proposed guaranteeing feasibility of the planned path when converted back into the original coordinate frame. The algorithm is posed as a convex optimization that takes into account the exact dimensions of the vehicle and the road, while minimizing the amount of overhang outside of the drivable surface.

The results of the proposed algorithm are compared in a simulation of a real road scenario against a centerline tracking scheme. The results show a significant decrease on the amount of overhang area outside of the drivable surface, leading to an increased safety in driving maneuvers. The real-time applicability of the method is shown, by using it in a receding-horizon framework.

13:00
Probabilistic controlled invariant set

ABSTRACT. This paper investigates stochastic invariance for stochastic control systems by controlled invariant sets (PCISs). We propose two denitions: N-step PCIS and innite-horizon PCIS, and provide necessary and sufficient conditions to verify that a set is PCIS.We also design iterative algorithms to compute the N-step PCIS and innite-horizon PCIS within a given set.

13:00
On estimating a nonlinear model of a quadcopter

ABSTRACT. In this work, we estimate a Hammerstein model of the vertical dynamics of a quadcopter. A nonlinear refined thrust model describing the relation between the calculated control signals for the rotors and the thrust is derived. The combination of this nonlinear model and closed-loop data poses some challenges and it is shown that an instrumental variables approach can be used to obtain accurate parameter estimates.

13:00
Quantifying the impact of cyber-attacks for control systems equipped with an anomaly detector

ABSTRACT. Risk assessment is an inevitable step in the implementation of cost-eective security strategies for control systems. One of the diculties of risk assessment is to estimate the impact cyber-attacks. This paper proposes a framework to estimate the impact of several cyber-attack strategies against a dynamical control system equipped with an anomaly detector. In particular, we consider denial of service, sign alternation, rerouting, replay, false data injection, and bias injection attack strategies. The anomaly detectors we consider are stateless, cumulative sum, and multivariate exponentially weighted moving average detectors. As a measure of the attack impact, we adopt the innity norm of critical states after a xed number of time steps. For this measure and the aforementioned anomaly detectors, we prove that the attack impact for all of the attack strategies can be reduced to the problem of solving a set of convex minimization problems, so the exact value of the attack impact can be obtained easily. We demonstrate how our modeling framework can be used for risk assessment on a numerical example.

13:00
Application of a linear PEM estimator to a stochastic Wiener-Hammerstein benchmark problem

ABSTRACT. In this contribution, we apply a computationally attractive consistent PEM estimator to a stochastic Wiener-Hammerstein benchmark problem using real-data. The estimator is defined using a one-step ahead predictor, which is linear in the observed outputs but nonlinear in the input. This allows us to tackle several issues considered challenging from the perspective of the current smoothing-based approaches.

13:00
Nonlinear FIR identification with model order reduction Steiglitz-McBride

ABSTRACT. In system identification, many structures and approaches have been proposed to deal with systems with non-linear behavior. When applicable, the prediction error method, analogously to the linear case, requires minimizing a cost function that is non-convex in general. The issue with non-convexity is more problematic for non-linear models, not only due to the increased complexity of the model, but also because methods to provide consistent initialization points may not be available for many model structures. In this paper, we consider a non-linear rational finite impulse response model. We observe how the prediction error method requires minimizing a non-convex cost function, and propose a three-step least-squares algorithm as an alternative procedure. This procedure is an extension of the Model Order Reduction Steiglitz-McBride method, which is asymptotically efficient in open loop for linear models. We perform a simulation study to illustrate the applicability and performance of the method, which suggests that it is asymptotically efficient.

13:00
Cloud application predictability through integrated load-balancing and service time control

ABSTRACT. Cloud computing provides the illusion of infinite capacity to application developers. However, data center provisioning is complex and it is still necessary to handle the risk of capacity shortages. To handle capacity shortages, graceful degradation techniques sacrifice user experience for predictability. In all these cases, the decision making policy that determines the degradation interferes with other decisions happening at the infrastructure level, like load-balancing choices. Here, we reconcile the two approaches, developing a load-balancing strategy that also handles capacity shortages and graceful degradation when necessary. The proposal is based on a sound control-theoretical approach. The design of the approach avoids the pitfalls of interfering control decisions. We describe the technique and provide evidence that it allows us to achieve higher performance in terms of emergency management and user experience.

13:00
Abstraction refinement for control synthesis: a discrete-time hybridization approach

ABSTRACT. This paper addresses the problem of synthesizing a controller for a nonlinear system to follow (in discrete-time) a sequence of regions in the partitioned state space. A discrete-time hybridization approach is first applied so that at each step of this sequence, the sampled nonlinear dynamics restricted to the considered partition cell are over-approximated by an affine abstraction. Mixed-integer linear programming (MILP) problems are then designed either to synthesize a controller for the affine model (and thus for the original system) to reach the next step of the sequence, or to detect infeasibility of this synthesis problem. When neither conclusion can be reached from solving the MILP problems, the affine abstraction is refined into a piecewise affine model by splitting the considered partition cell and repeating the hybridization step on each subcell. The control synthesis is then attempted again on the piecewise affine model of the refined abstraction. Once a controller is found for a cell, we proceed backwards on the sequence and apply the same method to the previous cell in the sequence.

13:00
Gait parameters learning for improved pedestrian dead reckoning algorithm

ABSTRACT. An improved Pedestrian Dead Reckoning (PDR) algorithm is proposed that is able to learn gait parameters whenever position estimates are available through a filtering approach. The estimated parameters obtained from multi-rate Kalman filter bank are used to improve the PDR algorithm in time intervals when the position measurements are unavailable.

13:00
Positioning in NB-IoT systems based on observed TDOA measurements

ABSTRACT. In this paper, we evaluate the potential of device tracking in narrowband internet of things (NB-IoT) systems using observed time difference of arrival (OTDOA) measurements. The reference signals for the timing measurement estimates are assumed to be reported from the user equipment (UE) to the location center periodically or on an on-demand basis. On-demand reporting based on the signal to noise ratio (SNR) of the measured cells received by the UE is simulated to investigate the possibility of optimizing the number of reports per minute budget to obtain the required horizontal positioning accuracy.

13:00
A Graph-theoretic approach to the H_{\infty} performance of dynamical systems on directed and undirected networks

ABSTRACT. We study a graph-theoretic approach to the $\mathcal{H}_{\infty}$ performance of leader following consensus dynamics on directed and undirected graphs. We first provide graph-theoretic bounds on the system $\mathcal{H}_{\infty}$ norm of the leader following dynamics and show the tightness of the proposed bounds. Then, we discuss the relation between the system $\mathcal{H}_{\infty}$ norm for directed and undirected networks for specific classes of graphs, i.e., balanced digraphs and directed trees. Moreover, we investigate the effects of adding directed edges to a directed tree on the resulting system $\mathcal{H}_{\infty}$ norm. At the end, we apply these theoretical results to a reference velocity tracking problem in a platoon of connected vehicles and discuss the effect of the location of the leading vehicle on the overall $\mathcal{H}_{\infty}$ performance of the system.

13:00
Residual based iterations and rational Krylov for generalized Lyapunov equations

ABSTRACT. Introduction For linear systems, Lyapunov equations play a key role in characterizing controllable and observable subspaces. By solving the Lyapunov equation one can compute the Gramians which can, e.g., be used in model order reduction, see the work of Antoulas (2005). For this end, efficient computational methods for large-scale problems have been developed over many years, for a recent summary see the work of Simoncini (2016).

Analogies can be made to, so called, bilinear systems of the form \begin{align}\label{eq:Bilsys} \dot x(t) = Ax(t) + \sum_{i=1}^m N_ix(t)u_i(t) + Bu(t), \end{align} where $x\in\mathbb{R}^n$, $u\in\mathbb{R}^m$, $A,N_i \in \mathbb{R}^{n\times n}$, and $B\in\mathbb{R}^{n\times m}$. For such a system, it has been shown in the work of Al-Baiyat and Bettayeb (1993), that the notion of Gramian can be generalized, and the controllability Gramian is given by the generalized Lyapunov equation, \begin{align}\label{eq:gen_Lyap} AX+XA^T+\sum_{i=1}^m N_iXN_i^T + BB^T = 0. \end{align} In analogy with the linear case, the bilinear Gramian gives information about controllability and is thus used to compute reduced order models, see the work of Benner and Damm (2011).

The generalized Lyapunov equation, \eqref{eq:gen_Lyap}, also arises naturally in other contexts such as stochastic bilinear control systems, and some partial differential equations on rectangular domains.

Problem formulation In this work we consider the Generalized Lyapunov equation, \eqref{eq:gen_Lyap}, which we view as the sum of a Lyapunov operator $\mathcal{L}(X) := AX+XA^T$ and a linear operator $\Pi(X) := \sum_{i=1}^m N_iXN_i^T$. We further assume that $\mathcal{L}$ is invertible, which happens for example if $A$ is Hurwitz, and that the (operator) spectral radius is $\rho(\mathcal{L}^{-1}\Pi)<1$. Under these assumptions it is guaranteed that the generalized Lyapunov equation \eqref{eq:gen_Lyap} has a solution.

For large-scale problems, computations can be intractable and the aim is to find efficient algorithms for computing low-rank approximations to \eqref{eq:gen_Lyap}.

Contribution The main contribution is two-fold. First we consider a method presented in the work of \Kressner and Sirkovic (2015). We show that their method, when applied to a generalized Lyapunov equation with $A$ symmetric and Hurwitz, $N_i$ symmetric, and with $\rho(\mathcal{L}^{-1}\Pi)<1$, is producing locally $\mathcal{H}_2$-optimal updates to the current approximation.

Secondly, we propose a new projection space for solving \eqref{eq:gen_Lyap}. Our method is based on a generalization of the Rational Krylov subspace method, and we propose to use the subspace \begin{align*} \mathcal{K}_k := \text{Span}\{B,(A-\sigma_1)^{-1}r_0,\dots,(A-\sigma_k)^{-1}r_{k-1}\}, \end{align*} where $r_k$ is the most dominant singular vector of the residual, i.e., $\mathcal{R}_k:=\mathcal{L}(\hat X_k) + \Pi(\hat X_k) + BB^T$, and where $\hat X_k$ is the current approximation in $\mathcal{K}_k$. A rational-Krylov-like method requires shifts to be selected, and we propose, in analogy with the work of Druskin and Simoncini (2011), to use \begin{align*} \sigma_{k+1} := \text{argmax}_{\sigma}\left(\left\| r_k - (A - \sigma I)\mathcal{V}_{k}(A_k - \sigma I)^{-1}\mathcal{V}_{k}^T r_k\right\|\right), \end{align*} where $\mathcal{V}_k$ is an orthogonal basis of $\mathcal{K}_k$, and $A_k = \mathcal{V}_k^TA\mathcal{V}_k$.

13:00
Discretization and tuning of nonlinear PID controllers

ABSTRACT. We address the problem of synthesizing controller gains in nonlinear PID control enabled by feedback linearization. Provided a linear structure in the error-dynamics typically arising from this approach, the algebraic Riccati equation is solved to provide optimal controller gain parameters, both in continuous and discrete time. The design methodology is showcased by tuning a non-linear PID controller for tracking control of rigid body dynamics on the SO(3).

13:00
State-dependent data queuing in shared-resource networked control systems

ABSTRACT. In the design of shared resource networked control systems (NCSs), resource managers play an important role to appropriately allocate limited resources across the distributed system. They are often used to fairly distribute the limited bandwidth among the medium-sharing entities at the expense of delaying or discarding unnecessary data samples. Considering the rapidly growing volume of information that need to be exchanged, a relevant scenario for efficient resource manage- ment is state-dependent data buffering through network queues. In this paper, we propose state-dependent data buffering for shared-resource NCSs, such that both buffer input and output can be adjusted depending on the real-time conditions of the control systems and the communication network. We consider that the transmission decisions at the sensor side are taken by event-based schedulers, and that those data eventually sent for transmission are queued and processed, depending on the available communication resource. To improve the buffer service, we propose a state-dependent buffer discharge method, based on real-time feedback from the communication network. We derive sufficient stability conditions under which the overall NCS with the proposed cross-layer transmission scheme is stable in mean-square sense. Moreover, we show performance improvements resulting from our proposed design in comparison with its state-independent counterpart.

13:00
A user-friendly implementation of the extended Rao-Blackwellized particle filter

ABSTRACT. We discuss how to implement the Rao-Blackwellized particle filter (RBPF) using only a nonlinear model (with derivatives of the dynamics and measurement functions available). The nonlinear model may be conditionally linear, or it may be approximated as such: the Extended RBPF (ERBPF). Any combination of linear and nonlinear measurement equations is supported. The resulting software implementation allows for easy usage of the \erbpf without extensive rewrites of the model code, for example when changing what states are approximated as linear or nonlinear. A linearized (EKF'') proposal is easily incorporated. The filter performance is demonstrated using a positioning problem.

13:00
A new method of scaling the gramian based input-output pairing methods for improved results

ABSTRACT. A key problem in the application of process control systems is to decide which inputs should control which outputs. There are multiple ways to solve this problem, among them using gramian based measures, which include the Hankel interaction index array, the participation matrix and the Σ2 method. These methods take into account system dynamics as opposed to many other methods which only consider the steady-state system. However, the gramian based methods have issues with input and output scaling. Generally, this is resolved by scaling all inputs and outputs to have equal range. We will, however, demonstrate how this can result in an incorrect pairing. Further, we examine scaling of the gramian based measures, using either row or column sums, or by utilizing the Sinkhorn-Knopp algorithm instead.

Then, to more systematically analyze the benefits of the scaling schemes, a multiple input multiple output model generator is used to test the different schemes on a large number of systems. This, along with implementation of automatic controller tuning, allows for a statistical comparison of the scaling methods. This assessment shows considerable benefits to be gained from the alternative scaling of the gramian based measures, especially when using the Sinkhorn-Knopp algorithm. The use of this method also has the advantage that the results are completely independent of the original scaling of the inputs and outputs.

13:00
Informative path planning with Gaussian target distributions

ABSTRACT. We study the problem of planning a path for a mobile sensor platform with limited field of view that is tracking a moving target. The state of the sensor is considered perfectly known and the state of the target is modeled by a Gaussian distribution, which makes the extended Kalman filter a suitable choice of estimator. The planning problem is formulated as a stochastic receding horizon control problem, where the stochasticity is caused by uncertainty in the target process noise and the measurement noise. We show that by approximating the expectation over the target state, a more robust plan is generated compared to when just the most likely target trajectory is considered.

13:00
A relative trajectory estimation algorithm based on orthogonal decomposition

ABSTRACT. In this research, we focus on the estimation problem of a small spacecraft in an astronomical small body flyby. We consider a 3-axis stabilized spacecraft with a rotating mirror that can change the line-of-sight direction at high speed in one axis direction. However, such estimation is difficult due to fast direction change. We propose a new estimation algorithm using orthogonal decomposition and a recursive least squares method. Our algorithm is based on linear operations and the information to be stored does not increase for each observation so that they can be applied to the embedded system with limited memory capacity and computational power.

13:00
On control structure design for a walking beam furnace

ABSTRACT. The aim of this article is to introduce a novel sparse controller design for the temperature control of an experimental walking beam furnace in steel industry. Adequate tracking of temperature references is essential for the quality of the heated slabs. However, the design of the temperature control is hindered by the multivariable (non-square) dynamic behavior of the furnace. These dynamics include significant loop interactions and time delays. Furthermore, a novel data-driven model, based on real life experimental data that relies on a subspace state representation in a closed loop approach is introduced. In the sequel, the derived model is utilized to investigate the controller’s structure. By applying the relative gain array approach a decentralized feedback controller is designed. However, in spite of the optimal and sparse design of the controller, there exists interaction between loops. By analyzing the interaction between the inputs-outputs with the Σ2 Gramian-based interaction methodology, a decoupled multi-variable controller is implied. The simulation result, based on the experimental modeling of the furnace, shows that the controller can successfully decrease the interaction between the loops and track the reference temperature set-points.

13:00
Dynamic models for the formal verification of big data applications via stochastic model checking

ABSTRACT. Big Data Applications (BDAs) manage so much data to require a cluster of machines for computation and storage. Their execution often has temporal constraints, such as deadlines to process the data. BDAs are executed within Big Data Frameworks(BDFs), that provide mechanisms to automatically manage the complexity of the computation distribution. For a BDA to fulfill its deadline when executed in a BDF, online dynamic resource allocation policies should be in place. The introduction of control for such resource allocation calls for formal verification of the closed-loop system. Model checkers verify the correct behaviour of programs, and in principle they could be used to prove properties on the BDF execution. However, the complexity of BDFs makes it infeasible to directly model the BDAs and BDFs. We propose a formalism to associate the execution of a BDA with a first-principle dynamic simulation model that can be used for model checking in the place of the real application, making the verification viable in practice. We introduce our formalism, apply it to a well assessed framework, and test its capabilities. We show that our solution is able to capture the dynamics and prove properties of the BDA execution using a stochastic model checker.

13:00
Tracking and sensor fusion in direction of arrival estimation using optimal mass transport

ABSTRACT. In this work, we propose new methods for information fusion and tracking in direction of arrival (DOA) estimation by utilizing optimal mass transport cost for comparing and interpolating spectra. Sensor array measurements in DOA estimation may not be consistent due to misalignments and calibration errors. We consider numerical examples where we use the proposed framework to prevent aliasing and correct array misalignments.

13:00
On event-based sampling for LQG-optimal control

ABSTRACT. We consider the problem of finding an event-based sampling scheme that optimizes the trade-off between average sampling rate and control performance in a linear-quadratic-Gaussian (LQG) control problem setting with output feedback. Our analysis is based on a recently presented sampled-data controller structure, which remains LQG-optimal for any choice of sampling scheme. We show that optimization of the sampling scheme is related to an elliptic convection–diffusion type partial differential equation over a domain with free boundary, a so called Stefan problem. A numerical method is presented to solve this problem for second order systems, and thus obtain an optimal sampling scheme. The method also directly generalizes to higher order systems, although with a higher computational cost. For the special case of multidimensional integrator systems, we present the optimal sampling scheme on closed form, and prove that it will always outperform its periodic counterpart. Tight bounds on the improvement are presented. The improved performance is also demonstrated in numerical examples, both for an integrator system and a more general case.

13:00
Compositional optimization of discrete event systems

ABSTRACT. Optimizing the execution of industrial processes such as manufacturing cells or whole assembly plants can have great impact on their performance. However, finding an optimal sequence of tasks in a large-scale system is a complex optimization problem. Most systems are comprised of multiple sub-systems and the search space of the optimization generally grows exponentially with each sub-system. In this paper, we propose the method compositional optimization to mitigate this problem. Compositional optimization integrates methods from optimization and compositional supervisory control theory to exploit the local behavior of the sub-systems, reducing them individually, and then synthesize a globally optimal controller compositionally. The local optimization technique avoids a monolithic model of the system, which can reduce the complexity of the optimization significantly. The potential of compositional optimization is demonstrated using a realistic example, similar to a large scale industrial application, while we also reflect on the limitations and highlight specific system properties that can be exploited by the method.

13:00
Developing concept inventory tests for Electrical Engineering (CITE): extractable information, early results, and learned lessons

ABSTRACT. This paper suggests a method for developing, implementing and assessing a concept inventory test for electrical engineering students (CITE). The aim of this test is to help students better understand and learn core concepts, plus increase their awareness about links between the different courses and other themes of the program. Our and other experiences show that students often struggle to understand and use fundamental concepts, and how these relate to the various courses. This issue is probably due to the fact that traditional exams mainly focus on assessing procedural tasks (e.g., directly solving specific problems following step-by-step approaches). The investigated programs at Uppsala University (UU) and Luleå University of Technology (LTU), nonetheless, have no tool for collecting quantitative data on how students develop conceptual knowledge throughout the programs, and thus no means to obtain an holistic view about their learning process. The here proposed methodology thus describes how to develop tests that would not only provide students with valuable feedback on their progression, but also equip teachers and program boards with high-end data for pedagogical and course development purposes. Besides illustrating the developmental methodology, the paper includes reactions and remarks from students on what the tests would provide and what would motivate them to take it.

13:00
An algorithm for computing explicit expressions for orthogonal projections onto subspaces of finite games

ABSTRACT. The space of finite games can be decomposed into three orthogonal subspaces, which are the subspaces of pure potential games, nonstrategic games, and pure harmonic games as shown by Candogan et al. (2011). This decomposition provides a systematic characterization for the space of finite games. Explicit expressions for the orthogonal projections onto the subspaces are helpful in analyzing general properties of finite games in the subspaces and the relationships of finite games in different subspaces. In the current paper, we give an algorithm for computing the explicit expressions only using the number of players.

13:00
Parallel exploitation for tree-structured coupled QP solver in Julia

ABSTRACT. The main idea in this work is to implement a distributed primal-dual interior-point algorithm for loosely coupled Quadratic Programming problems. We implement this in Julia and exploit parallelism in order to increase computational speed. We benchmark performance of the algorithm on a Model Predictive Control problem.

13:00
Learning localized spatio-temporal models from streaming data

ABSTRACT. We address the problem of predicting spatio-temporal processes with temporal patterns that vary across spatial regions, when data is obtained as a stream. That is, when the training dataset is augmented sequentially. Specifically, we develop a localized spatio-temporal covariance model of the process that can capture spatially varying temporal periodicities in the data. We then apply a covariance-fitting methodology to learn the model parameters which yields a predictor that can be updated sequentially with each new data point. The proposed method is evaluated using both synthetic and real climate data which demonstrate its ability to accurately predict data missing in spatial regions over time.

13:00
Single-pass observation update of smoothing posterior

ABSTRACT. A method is derived for updating the smoothing posterior of a linear Gaussian state-space model in a single pass at the reception of a new observation. Commonly the posterior distribution would be computed from scratch using a two-pass formula, such as the Rauch-Tung-Striebel smoother.

13:00
How consistent is my model with the data? Information-theoretic model check

ABSTRACT. The choice of model class is fundamental in statistical learning and system identification, no matter whether the class is derived from physical principles or is a generic black-box. We develop a method to evaluate the specified model class by assessing its capability of reproducing data that is similar to the observed data record. This model check is based on the information-theoretic properties of models viewed as data generators and is applicable to e.g. sequential data and nonlinear dynamical models. The method can be understood as a specific two-sided posterior predictive test. We apply the information-theoretic model check to both synthetic and real data and compare it with a classical whiteness test.

15:00-17:00 Session 4A: Mobile robots
Location: Q1
15:00
The Electrolux Pure i9 robotic vacuum cleaner - a peek under the hood

ABSTRACT. Electrolux recently launched the Pure i9 robotic vacuum cleaner, which uses a novel 3D laser scanner to avoid obstacles and do SLAM. This talk will describe the sensor and the core algorithms used for SLAM, obstacle avoidance, path planning and cleaning patterns

15:20
Towards autonomous surveying of underground mine using MAVs

ABSTRACT. Micro Aerial Vehicles (MAVs) are platforms that received great attention during the last decade. Recently, the mining industry has been considering the usage of aerial autonomous platforms in their processes. This article initially investigates potential application scenarios for this technology in mining. Moreover, one of the main tasks refer to surveillance and maintenance of infrastructure assets. Employing these robots for underground surveillance processes of areas like shafts, tunnels or large voids after blasting, requires among others the development of elaborate navigation modules. This paper proposes a method to assist the navigation capabilities of MAVs in challenging mine environments, like tunnels and vertical shafts. The proposed method considers the use of Potential Fields method, tailored to implement a sense-and-avoid system using a minimal ultrasound-based sensory system. Simulation results demonstrate the effectiveness of the proposed strategy.

15:40
Combining homotopy methods and numerical optimal control to solve motion planning problems

ABSTRACT. This work presents a systematic approach for computing local solutions to motion planning problems in non-convex environments using numerical optimal control techniques. It extends the range of use of state-of-the-art numerical optimal control tools to problem classes where these tools have previously not been applicable. Today these problems are typically solved using motion planners based on randomized or graph search. The general principle is to define a homotopy that transforms, or preferably relaxes, the original problem to an easily solved problem. The approach is demonstrated in motion planning problems in challenging 2D and 3D environments, where the presented method significantly outperforms both a state-of-the-art numerical optimal control method and a state-of-the-art open-source optimizing sampling-based planner commonly used as benchmark.

16:00
Energy and peak-power optimization of existing time-optimal robot trajectories

ABSTRACT. This paper presents an optimization procedure that reduces up to 30% of energy consumption and up to 60% of peak-power for the trajectories that were tested on a real industrial robot. We have evaluated a number of cost functions and examined our method for a variety of scenarios such as varying cycle times and single/two-robot cases. The significance of our work is not only in the impressive savings, the simplicity of implementation and preserving path and cycle time but also in the effort made to carry out the optimization and experiments in as realistic conditions as possible.

16:20
Modelling and control of a tilt-wing unmanned aerial vehicle

ABSTRACT. In this article a Tilt-Wing Unmanned Aerial Vehicle (TW-UAV) and the preliminary evaluation of its hovering characteristics in extended simulation studies are presented. In the beginning, an overview of the TW-UAV's design properties are established, highlighting the novelties of the proposed structure and the overall merits. The TW-UAV's design and structural properties are mathematically modeled and utilized for the synthesis of a cascaded P-PI and PID based control structure for the regulation of its hovering performance. In addition, extensive simulation trials are performed in order to evaluate the structure's efficiency in controlling the TW-UAV's attitude and position under various noise and disturbance scenarios.

16:40
Detection and control of contact force transients in robotic manipulation without a force sensor

ABSTRACT. In this research, it is shown that robot joint torques can be used to recognize contact force transients induced during robotic manipulation, thus detecting when a task is completed. The approach does not assume any external sensor, which is a benefit compared to the state of the art. The joint torque data are used as input to a recurrent neural network (RNN), and the output of the RNN indicates whether the task is completed. A real-time application for force transient detection is developed, and verified experimentally on an industrial robot.

15:00-17:00 Session 4B: Transportation systems
Location: Q2
15:00
Model predictive supervisory control for in-use emission compliance of heavy-duty vehicles

ABSTRACT. This paper presents an MPC based strategy for integrated control of the engine and the exhaust aftertreatment system (EATS) of a heavy-duty powertrain. The objectives of this control strategy are to improve the fuel economy while fulfilling in-use emission legislation based on work-based-window, as stipulated in the Euro VI legislation. The strategy is evaluated using a high-fidelity simulation platform including a validated GT-POWER model of a 13L turbo compound diesel engine and a first principles model of an EATS. Simulation results show that the strategy always fulfills the in-use emission legislation.

15:20
Decentralized traffic light control with dynamic cycle length

ABSTRACT. An established rule of thumb in the field of traffic light control prescribes that, during periods of higher demand, it is convenient to have longer cycles. This is in order to reduce the fraction of the cycle length when no incoming lanes receive green light when the queues are long. In this extended abstract, we present a novel, provably stable, decentralized feedback traffic light control policy with variable cycle length. The proposed control strategy is fully decentralized and does not require any information about the network structure or the turning rates. Simulations in the micro-simulator SUMO show that having dynamic cycle lengths allows one to significantly reduce the overall queue lengths in the network, in both medium and low demands.

15:40
A safety zone based approach for effective and secure traffic control at intersections

ABSTRACT. In this paper, an optimization based algorithm for safe and efficient colaborative driving in intersections is extended and reformulated. The problem is to determine the optimal order in which vehicles should travel through a crossing under the assumptions that we can control all vehicles lateral velocity along predefined path and can determine in which order they should move through the crossing. In the original formulation one quadratic optimization program was solved for each possible crossing order of the vehicles and collisions were avoided by formulating constraints that only allowed one vehicle at the time, in the intersection. To make this algorithm more effective, we propose to formulate less restrictive collision avoidance constraints by introducing a number of critical zones based on where the predefined paths cross each other. We show that in this new formulation lead to a decrease in the number of quadratic optimization programs that need to be solved to find the optimal solution. Further, we provide an algorithm which guarantee that guarantees to find the minimum number of crossing sequences that need to be explored to guarantee the best solution. The results show that when simulating more complex scenarios, like four vehicles travelling through an ordinary crossing, the reduction of computational time and the time it takes for all vehicles to make it through the crossing can be significantly reduced by using the extensions proposed in this paper.

16:00
On the convergence of driver centric zone pricing for traffic networks

ABSTRACT. Congestion fees with dynamical zone pricing provide an easy-to-implement solution to urban traffic control agencies for improved efficiency of the traffic network. In order to design a pricing strategy it is necessary to understand how it affects the route choices of drivers in the system. This work explores this question by providing an idealized dynamical model for re-routing as traffic flows change to cheaper routes. Results show a convergence to a Wardrop-equilibrium, which is also proven formally and demonstrated via simulation.

16:20
Safety analysis for controller handover in mobile systems

ABSTRACT. Next generation mobile networks are envisioned to provide support for real-time control applications. One of the main aspects of these systems is that the location of the controller may be separated from the location of sensing and actuation. This promises benefits in terms of an increased flexibility, lower costs due to resource sharing, and higher computational capabilities. This paper focuses on one aspect of such systems, specifically, the controller handover. During a controller handover, a control process is moved from one point of computation to another at runtime. A possible reason for performing such a handover is to move the control process to a controller with better channel conditions. The safety of the handover is analyzed using a probabilistic reachability analysis by modeling the handover procedure as a stochastic hybrid system. Based on this safety analysis, a safety-oriented handover triggering rule is proposed. This triggering rule is shown to be dependent on the instantaneous state of the plant, in contrast to handover in mobile networks where it is only dependent on the state of the communication links. A vehicle platoon is considered as an example scenario, which is controlled by a base station of a mobile network. While driving, the platoon will move out of the communication range of the base station, so the control process needs to be moved to the next base station. Simulations illustrate the conditions for a safe execution of so called hard and soft handover protocols.

16:40
Experimental validation of a distributed MPC strategy for automated cars at intersections

ABSTRACT. In this paper we discuss a Model Predictive Control (MPC) strategy for automated vehicles at intersections. The MPC consists of an outer loop, which provides non-overlapping intersection occupancy timeslots, and inner control loops, which gives the vehicle control commands. The timeslots are obtained as the solution to a Nonlinear Program (NLP). We present data from an experimental validation of the MPC strategy in which the NLP was solved in a distributed fashion over vehicle-to-vehicle communication links using a tailored Sequential Quadratic Programming Algorithm. The results indicate that the MPC strategy is practically relevant and able to perform safe and efficient coordination.

15:00-17:00 Session 4C: System identification
Location: Q31
15:00
On model selection for high dimensional linear regressions

ABSTRACT. Model selection is a great challenge when the number of measurements is much smaller than the dimension of the parameter space. We study the problem of regressor selection for high-dimensional linear regressions. We propose a new model selection criterion that leads to the selection of a parsimonious model from all the combinatorial models up to some maximum model order. Under given assumptions we show that the proposed criterion provides the true model with a probability approaching one as the number of measurements tends to infinity, or when the noise variance tends to zero. To practically use the criterion, we propose to combine it with the LARS/LASSO estimator for sparse linear regression. The proposed criterion is then used to choose the regularization parameter for the LASSO estimator such that it selects the true variables.

15:20
Convex bounds for equation error in stable nonlinear identification

ABSTRACT. Equation error, a.k.a. one-step-ahead prediction error, is a common quality-of-fit metric in dynamical system identification and learning. In this paper, we use Lagrangian relaxation to construct a convex upper bound on equation error that can be optimized over a convex parametrization models with guaranteed stability. We provide theoretical results on the tightness of the relaxation, and show that the method compares favourably to established methods on a variety of case studies.

15:40
An asymptotically optimal indirect approach to continuous-time system identification

ABSTRACT. The indirect approach to continuous-time system identification consists in estimating continuous-time models by first determining an appropriate discrete-time model. For a zero-order hold sampling mechanism, this approach usually leads to a transfer function estimate with relative degree 1, independent of the relative degree of the strictly proper real system. In this paper, a refinement of these methods is developed. Inspired by indirect PEM, we propose a method that enforces a fixed relative degree in the continuous-time transfer function estimate, and show that the resulting estimator is consistent and asymptotically efficient. Extensive numerical simulations are put forward to show the performance of this estimator when contrasted with other indirect and direct methods for continuous-time system identification.

16:00
Closed-Loop identification using instruments simulated from nonlinear models

ABSTRACT. In this contribution possible benefits and drawbacks of using nonlinear models to simulate instruments in the refined Instrumental variable (IV) method are observed. This is done by comparing the results with those of the more commonly used approach, to use high order autoregressive models with exogenous inputs (ARX). It is difficult to find a universal nonlinear structure that always gives good results. A too complex structure will quickly overfit to noise whereas a too simple structure might not be able to explain the system at hand. Simulations indicate that, with carefully selected regularization, it is possible to outperform linear IV identification, with auxiliary ARX models, quite often.

16:20
Regularized parametric system identification: a decision-theoretic formulation

ABSTRACT. Parametric prediction error methods constitute a classical approach to the identification of linear dynamic systems with excellent large-sample properties. A more recent regularized approach, inspired by machine learning and Bayesian methods, has also gained attention. Methods based on this approach estimate the system impulse response with excellent small-sample properties. In several applications, however, it is desirable to obtain a compact representation of the system in the form of a parametric model. By viewing the identification of such models as a decision, we develop a decision-theoretic formulation of the parametric system identification problem that bridges the gap between the classical and regularized approaches above. Using the output-error model class as an illustration, we show that this decision-theoretic approach leads to a regularized method that is robust to small sample-sizes as well as overparameterization.

16:40
Parametric system identication with least-squares methods

ABSTRACT. In identification of parametric models, the prediction error method using a quadratic cost function provides asymptotically efficient estimates under Gaussian noise and additional mild assumptions, but in general it requires solving a non-convex optimization problem. Two other important classes of methods are instrumental variable and subspace methods, which do not suffer from non-convexity but are in general not asymptotically efficient.

Other methods estimate an intermediate non-parametric model, which can then be used to obtain the parametric model of interest. Recent developments have used least squares for this purpose, with weighting or preceded by a filtering step. In particular, the Weighted Null-Space Fitting and the Model Order Reduction Steiglitz-McBride methods are asymptotically efficient under similar assumptions as PEM, while not explicitly minimizing a cost function. These methods are also quite flexible in parametrization, allowing for several extensions. Examples are identification of unstable systems, dynamic networks, multi-input multi-output systems, and recursive identification.

In this presentation, we perform a survey of the recent advancements in parametric identication using least-squares methods. We cover the basic methods and extensions, pointing out advantages, limitations, and future directions.

15:00-17:00 Session 4D: Networked systems I
Location: Q34
15:00
Formal guarantees and evaluation of a event-driven bandwidth allocation scheme for camera networks

ABSTRACT. Modern computing systems in which a multitude of devices compete for network resources suffer from performance issues derived from inefficient bandwidth allocation policies. This problem is often mitigated in bandwidth-constrained systems by introducing run-time device-level adaptations (e.g. adjustment of operation parameters) to ensure correct information transmission. Adapting their behavior, the devices are capable of consequently adjusting their bandwidth requirements. Such scenarios present two main issues: (i) the adaptation at the device level can interfere with network allocation policies; (ii) it is quite difficult to obtain formal guarantees on the system’s behavior, given that multiple adaptation strategies (network distribution and device-level adaptation) are active at the same time – and hence the presence of multiple independent control loops may lead to interference and result in disruptive effects. In this work, we tackle the two aforementioned issues in bandwidth allocation, ensuring the satisfaction of formal properties like convergence to a steady state using model checking. We apply our method to a camera surveillance network, in which self-adaptive cameras compete for network resources to send streams of frames to a central node.

15:20
Time-constrained multi-agent task scheduling

ABSTRACT. The problem of time-constrained multi-agent task scheduling is addressed. For a group of agents, we assume the existence of a set of dynamic activating tasks, each of which is associated with a relative deadline and several Quality-of-Service levels. By taking into account the reward and cost of satisfying the tasks, a dynamic scheduling strategy is proposed.

15:40
Resource control of a network of smart services hosting automation applications

ABSTRACT. To remain competitive in the field of manufacturing today, companies must make their industrial robots smarter and allow them to collaborate with one another in a more effective manner. This is typically done by adding some form of learning, or artificial intelligence (AI), to the robots. These learning algorithms are often packaged as cloud functions in a remote computational center since the amount of computational power they require is unfeasible to have at the same physical location as the robots.

In this work we address this challenge by deriving a rigorous mathematical framework that models a general network of cloud functions. On top of this network several applications are hosted. Using this framework we propose a generalized AutoSAC (automatic service- and admission controller) that builds on previous work by the authors. In the previous work the system was only capable of handling a single set of cloud functions, with a single application hosted on top of it. With the contributions of this paper it becomes possible to host multiple applications on top of a larger, general network of cloud functions. It also allows for each application to have its own end-to-end deadline requirement.

16:00
Privacy-aware minimum error probability estimation: an entropy constrained approach

ABSTRACT. This paper studies the design of an optimal privacy-aware estimator for a single sensor estimation problem. The sensor's measurement is a (possibly non-linear) function of a private random variable, a public random variable and the measurement noise. Both public and private random variables are assumed to be discrete valued, and the measurement noise is arbitrarily distributed. The sensor provides an estimate of the public random variable for an untrusted entity, named the cloud. The objective is to design the estimator of the public random variable such that a level of privacy for the private random variable is guaranteed. The privacy metric is defined as the discrete conditional entropy of the private random variable given the output of the estimator. A binary loss function is considered for the estimation of the public random variable. The optimal estimator design problem is posed as the minimization of the average loss function subject to a constraint on the privacy level of the private random variable. It is shown that the objective function is linear and the privacy constraint is convex in the optimization variables. Thus, the optimal privacy-aware estimator can be designed by solving an infinite dimensional convex optimization problem.

16:20
Collective decision-making in multiagent networks

ABSTRACT. For a particular class of nonlinear interconnected systems, where the nonlinearities are sigmoidal and saturated, we investigate how the frustration of the social network affects the decision-making process in a community of agents, where both cooperative and antagonistic interactions coexist. It is shown that the value of social commitment required from the agents in order to achieve a nontrivial collective decision grows with the frustration of the network.

16:40
Multi-agent formation control for circumnavigation of dynamic shapes

ABSTRACT. The problem of multi-agent formation control for target tracking is considered in this paper. The target is an irregular dynamic shape approximated by a circle with moving centre and varying radius. It is assumed that there are n agents and one of them is capable of measuring both the distance to the boundary of the target and to its centre. All the agents must circumnavigate the boundary of the target while forming a regular polygon. Protocols are designed for two cases: limit bounded tracking with no actuator disturbances and finite time tracking considering actuator disturbances. One simulated example is provided to verify the usage of one of the control protocols designed in this paper.