Séminaire d'Automatique du Plateau de Saclay : Distributed Abstractions for Multi-Agent Systems Based on Robust Multi-Agent Control

Séminaire le 7 Novembre 2017, 10h00 à CentraleSupelec (Gif-sur-Yvette) Salle des séminaires du L2S
Dimitris Boskos (Postdoctoral researcher, Department of Automatic Control, School of Electrical Engineering, Royal Institute of Technology (KTH), Stockholm, Sweden)

High level task planning for multi-agent systems constitutes a research area which has gained an emerging attention during the last two decades. While the agents' coordination is in principle based on the design of continuous interaction protocols, the derivation of high level plans requires a discrete representation of their dynamic behavior, also called abstraction, in order to leverage algorithmic tools for the plan synthesis.    

In this talk we discuss the derivation of such abstractions for agents with continuous dynamics, comprising of feedback interconnection terms and additive bounded inputs, which provide the ability for high level planning under the coupled constraints. These dynamics are also motivated by multi-agent coordination protocols which are robust with respect to the additional input part. We will present such a cooperative control framework, which guarantees that network connectivity is robustly maintained with respect to bounded additive inputs. Furthermore, a modification of the feedback design ensures forward invariance of the agents' trajectories inside a convex workspace, without affecting the inputs' robustness bounds.

In order to derive the agents' distributed symbolic models, we determine space-time discretizations which establish that each agent's abstraction has at least one outgoing transition from every discrete state. The symbolic model of each agent is based on the knowledge of its neighbors' discrete positions and the transitions are performed through hybrid control laws, which can drive the agent to its possible successor states. As an extension of these results we also consider a varying degree of decentralization and build each abstract model based on discrete information up to a tunable distance in the communication graph. Finally, we discuss the derivation of online  abstractions, by discretizing over approximations of the agents' reachable sets over a bounded time horizon.

Bio. Dimitris Boskos was born in Athens, Greece in 1981. He has received the Diploma in Mechanical Engineering from the National Technical University of Athens (NTUA), Greece, in 2005, the M.Sc. in Applied Mathematics from the NTUA in 2008 and the Ph.D. in Applied mathematics from the NTUA in 2014. Since August 2014, he is a Postdoctoral Researcher at the Department of Automatic Control, School of Electrical Engineering, Royal Institute of Technology (KTH), Stockholm, Sweden. His research interests include distributed control of multi-agent systems, formal verification and observer design for nonlinear systems.