DECIDE - AI design of decentralized coopetitive control over networks

The project has resulted in the publications listed below. Use the "»" button to reveal an abstract of each publication, and "«" to hide the abstract again (requires Javascript).

Journal papers

T. Santejudean, S. Ungur, R. Herzal, I.-C. Morarescu, V.S. Varma, L. Busoniu, Globally convergent path-aware optimization with mobile robots. Nonlinear Analysis - Hybrid Systems, vol. 55, pages/article number 101546, 2025. »
Abstract: Consider a mobile robot that must navigate as quickly as possible to the global maxima of a function (e.g. density of seabed litter, pollutant concentration, wireless signal strength) defined over its operating area. This objective function is initially unknown and is assumed to be Lipschitz continuous. The limited velocity of the robot restricts the next samples to neighboring positions, and to avoid wasting time and energy, the robot's path must be adapted as new information becomes available. The paper proposes two methods that use an upper bound on the objective to iteratively change the position targeted by the robot as new samples are acquired. The first method is FTW, which Turns When the best value seen so far of the objective Function is larger than the bound of the current target position. The second is FTWD, an extension of FTW that takes into account the Distance to the target. Convergence guarantees are provided for both methods, and a convergence rate is proven to characterize how fast the FTW suboptimality decreases as the number of samples grows. In a numerical study, FTWD greatly improves performance compared to FTW, outperforms two representative source-seeking baselines, and obtains results similar to a much more computationally intensive method that does not guarantee convergence. The relationship between FTW and FTWD is also confirmed in real-robot experiments, where a TurtleBot3 seeks the darkest point on a 2D grayscale map.
Online at ScienceDirect.
«
B. Yousuf, R. Herzal, Zs. Lendek, L. Busoniu, Multi-agent active multi-target search with intermittent measurements. Control Engineering Practice, vol. 153, pages/article number 106094, 2024. »
Abstract: Consider a multi-agent system that must find an unknown number of static targets at unknown locations as quickly as possible. To estimate the number and positions of targets from noisy and sometimes missing measurements, we use a customized particle-based probability hypothesis density filter. Novel methods are introduced that select waypoints for the agents in a decoupled manner from taking measurements, which allows optimizing over waypoints arbitrarily far in the environment while taking as many measurements as necessary along the way. Optimization involves control cost, target refinement, and exploration of the environment. Measurements are taken either periodically, or only when they are expected to improve target detection, in an event-triggered manner. All this is done in 2D and 3D environments, for a single agent as well as for multiple homogeneous or heterogeneous agents, leading to a comprehensive framework for (Multi-Agent) Active target Search with Intermittent measurements (MA)ASI. In simulations and real-life experiments involving a Parrot Mambo drone and a TurtleBot3 ground robot, the novel framework works better than baselines including lawnmowers, mutual-information-based methods, active search methods, and our earlier exploration-based techniques.
Online at Elsevier.
«
A. Couthures, V.S. Varma, S. Lasaulce, I.-C. Morarescu, Analysis of an opinion dynamics model coupled with an external environmental dynamics. Chaos, Solitons & Fractals, vol. 189, pages/article number 115719, 2024. »
Abstract: We consider a set of individuals, referred to as agents, whose opinions evolve according to nonlinear dynamics. Their opinions impact their behavior or actions, which in turn affect their local environment (for example, via pollution, contamination of a virus, etc.). Each agent can also perceive or observe a signal about the environment, and is influenced by this external signal. This yields a coupled dynamics (opinion and external signal), which behaves in a similar manner to the preypredator models. One of the main features of our study is that the information provided by the external signal has a significant impact on the opinion dynamics. When the coupling is strong, the external signal may induce either chaotic behavior or convergence towards a limit cycle. When the coupling with the external signal is weak, the classical behavior characterized by local agreements in polarized clusters is observed. In both cases, conditions under which clusters of individuals do not change their actions are provided. Numerical examples are provided to illustrate the derived analytical results.
Online at Elsevier.
«
D. R. Alkhorshid, E. S. Tognetti, I.-C. Morarescu, Saturated control of consensus value under energy and state constraints in multi-agent systems. Automatica, vol. 69, pages/article number 111822, 2024. »
Abstract: This work presents a novel decentralized control strategy with a guaranteed cost for bilinear multiagent systems subjected to products between the state and the control input, state constraints, and limitations on the amplitude and total energy of the control action, which can prevent the consensus from reaching the desired value. We propose state feedback and switching control laws to deal with these restrictions. The main objectives of the work are twofold: (i) to design control laws that ensure stability and guaranteed cost bounds under constraints, and (ii) to determine an estimation of the domain of attraction (DOA), as large as possible, characterized by a polyhedral and an ellipsoidal invariant region contained in the space defined by the state constraints. We adopt a convex optimization procedure based on linear matrix inequalities (LMI) to address these objectives. An original approach employing Lyapunov sets is proposed to deal with control energy and state constraints, and positive system properties are used to estimate the DOA in only one orthant of state space. Through numerical examples, we demonstrate the effectiveness of the proposed Lyapunov-based approach, showing its ability to handle complex constraints and large networks.
Online at Elsevier.
«
O. Lindamulage De Silva, S. Lasaulce, I.-C. Morarescu, V. S. Varma, A game theory analysis of decentralized epidemic management with opinion dynamics. IEEE Transactions on Control of Networked Systems, 2024. Early access. »
Abstract: In this paper, we introduce a static game that allows one to numerically assess the loss of efficiency induced by decentralized control or management of a global epidemic. Each player represents a region, which is assumed to choose its control to implement a tradeoff between socio-economic aspects and health aspects; the control comprises both epidemic control physical measures and influence actions on the region's opinion. The Generalized Nash equilibrium (GNE) analysis of the proposed game model is conducted. The direct analysis of this game of practical interest is non-trivial but it turns out that one can construct an auxiliary game which allows one: to prove existence and uniqueness; to compute the GNE and the optimal centralized solution (sum-cost) of the game. These results allow us to assess numerically the loss (measured in terms of Price of Anarchy (PoA) induced by decentralization with or without taking into account the opinion dynamics.
Online at IEEEXplore.
«
P. Lorenzetti, M. Giaccagli , I.-C. Morarescu, R. Postoyan, Singularly Perturbed k-Contractive Linear Systems. IEEE Control Systems Letters, vol. 8, pages/article number 1144-1149, 2024. »
Abstract: A dynamical system is said to be k-contractive when its trajectories contract k-dimensional volumes. For k=1 , this property coincides with the classical notion of contraction. However, for k>1, it allows to characterize a much richer asymptotic behavior. The property of k-contraction has been introduced only recently, thus many analysis tools that are key in relevant applications are currently lacking for k-contractive systems. Motivated by this, we study k-contraction for singularly perturbed systems (SPSs), which naturally arise in many engineering applications. In particular, we focus on singularly perturbed linear time-invariant (LTI) systems. First we show that, for a 'sufficiently large' time-scale separation, the k-contraction properties of a SPS can be derived from those of the associated boundary-layer (fast) system, and from the dimension of the reduced order (slow) model. Then, we focus on the case in which the reduced order (slow) model is k-contractive and the boundary-layer (fast) system is 1-contractive. In this setting, we provide a stronger result by showing that the overall system is k-contractive when the time-scale separation is 'large'.
Online at IEEEXplore.
«
M. Susca, V. Mihaly, Zs. Lendek, I.-C. Morarescu, Passivity of Linear Singularly Perturbed Systems. IEEE Control Systems Letters, vol. 8, pages/article number 2105-2110, 2024. »
Abstract: The passivity of singularly perturbed systems (SPSs) is generally studied without taking advantage of the time-scale separation present in this class of systems. To fill this gap, the objective of this letter is to provide easy-to-verify well-posed conditions characterizing the passivity of a perturbation variable-dependent SPS starting from the passivity of its associated reduced-order system. To achieve this goal, we rely on the connection between positive realness and passivity, as well as the notion of phase for multi-input multi-output (MIMO) systems. We use a benchmark DC motor to illustrate that classical reasoning used for stability analysis of SPSs, which is based on the stability of the reduced-order (slow) and boundary layer (fast) subsystems, cannot be applied to guarantee the passivity of an SPS. On top of that, our methodology explains how the time-scale separation can be used to analyze the passivity of general linear time-invariant (LTI) systems. The approach is illustrated on a numerical example.
Online at IEEEXplore.
«
B. Yousuf, Zs. Lendek, L. Busoniu, Exploration-Based Planning for Multiple-Target Search with Real-Drone Results. Sensors, vol. 24, pages/article number 2868, 2024. »
Abstract: Consider a drone that aims to find an unknown number of static targets at unknown positions as quickly as possible. A multi-target particle filter uses imperfect measurements of the target positions to update an intensity function that represents the expected number of targets. We propose a novel receding-horizon planner that selects the next position of the drone by maximizing an objective that combines exploration and target refinement. Confidently localized targets are saved and removed from consideration along with their future measurements. A controller with an obstacle-avoidance component is used to reach the desired waypoints. We demonstrate the performance of our approach through a series of simulations as well as via a real-robot experiment in which a Parrot Mambo drone searches from a constant altitude for targets located on the floor. Target measurements are obtained on-board the drone using segmentation in the camera image, while planning is done off-board. The sensor model is adapted to the application. Both in the simulations and in the experiments, the novel framework works better than the lawnmower and active-search baselines.
Online at MDPI.
«
B. Adhikari, J. Veetaseveera, V.S. Varma, I.-C. Morarescu, E. Panteley, Computationally efficient guaranteed cost control design for homogeneous clustered networks. Automatica, vol. 163, pages/article number 111588, 2023. »
Abstract: We consider a clustered network where connections inside the cluster are dense and between clusters are sparse. This leads us to a classical decoupling into fast (intra-cluster) and slow (inter-cluster) dynamics. Our objective is to provide a computationally efficient method to design control strategies that guarantee a certain bound on the cost for each cluster. Basically, we design a composite synchronizing controller with two terms: one responsible for the intra-cluster synchronization and the other achieving the synchronization between clusters. The first one does not require much computational effort since an analytic expression describes it. The second term is designed through a satisfaction equilibrium approach. In other words, the internal (fast) and external (slow) controllers are independently designed, and they ensure a guaranteed satisfactory cost for each cluster. Moreover, we show that the internal control affects the cluster cost only for a short time period. Finally, numerical simulations illustrate the theoretical results.
Online at ScienceDirect.
«

Conference papers

M. Giaccagli, V.S. Varma, D. Astolfi, Linear output regulation of discrete-time networked systems subject to stochastic packet drops. To appear in 2024 IEEE 63rd Conference on Decision and Control (CDC), Milan, Italy, 16–19 December 2024. »
Abstract: We investigate a scenario in which a given discrete-time controller communicates with a discrete-time plant via a wireless erasure channel in a linear scenario. The controller is designed for the output of the plant to track a periodic reference generated by a finite superimposition of discretetime linear oscillators while rejecting additional disturbances, i.e., to solve an output regulation problem. Due to stochastic packet drop induced by the network, the closed-loop behavior switches between two different dynamical systems, depending on the probability of transmission. We show that, when packet drops occur on the channel from the controller to the actuator, the exogenous signal implies that the expected regulation error does not go to zero but converges to a ball centered at zero. Differently, when the packet drops are on the output channel, we provide a set of sufficient conditions such that the regulation error asymptotically shrinks to zero in expectation. Results are validated via numerical examples. «
E. Pop, I. Lal, L. Busoniu, Real-Time Simultaneous Optimistic Planning for Hybrid-Input Nonlinear Optimal Control. In 2024 IEEE International Conference on Automation, Quality and Testing, Robotics (AQTR), Cluj-Napoca, Romania, 16–18 May 2024. »
Abstract: Simultaneous Optimistic Planning for Hybrid-Input Systems (SOPHIS) is a powerful method for the near-optimal control of nonlinear systems with hybrid - continuous and discrete - inputs, which works by iteratively splitting sets of input sequences. The generality of SOPHIS however comes at high computational costs that are often untenable in real-time control, especially for fast unstable systems. We introduce two modifications that make SOPHIS more suitable for real-time control: running it on a separate machine, over multiple sampling periods, while applying several inputs to the system during this time; and parallelizing the algorithm by splitting several sets simultaneously across multiple threads. Experiments investigate two parallelization schemes, the impact of thread count on the execution time, and the influence of the prediction horizon and budget; the latter on a real-life fast unstable system, a rotary inverted pendulum. In the experiments, the discrete input controls the quantization accuracy of the control action sent to the system.
Online at IEEEXplore.
«
A. Couthures, T. Mongaillard, V. S. Varma, S. Lasaulce, I.-C. Morarescu, Analysis of a continuous opinion and discrete action model coupled with an external dynamics. In 22nd European Control Conference (ECC), Stockholm, Sweden, 25–28 June 2024. »
Abstract: We consider a set of consumers in a city or town whose opinion is governed by a continuous opinion and discrete action (CODA) dynamics model. This dynamics is coupled with an observation signal dynamics, which defines the information on the common pollution that the consumers can access. We show that the external observation signal has a significant impact on the asymptotic behavior of the CODA model. When the coupling is strong, it induces either a chaotic behavior or convergence towards a limit cycle. When the coupling is weak, a more classical behavior characterized by local agreements in polarized clusters is observed. In both cases, conditions under which clusters of consumers dont change their actions are provided. Numerical examples are provided to illustrate the derived analytical results.
Online at IEEEXplore.
«

Disclaimer: The following applies to the papers that are directly available for download as PDF files. This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each copyright holder. In most cases, these works may not be reposted without the explicit permission of the copyright holder. Additionally, the following applies to IEEE material: Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works, must be obtained from the IEEE.