Robust and Decentralized Control of Multi-agent Systems under High-level Tasks
Time: Fri 2019-12-06 09.00
Subject area: Electrical Engineering
Doctoral student: Alexandros Nikou , Reglerteknik, Division of Decision and Control Systems
Opponent: Michael Zavlanos,
Supervisor: Dimos V. Dimarogonas, Reglerteknik
Decentralized control of multi-agent systems is an active topic of research, with many practical applications arising in multi-robot systems, autonomous driving, transportation systems and robotic manipulation. The contributions of this thesis lie in the scope of three topics: formation control, robust decentralized tube-based nonlinear Model Predictive Control and time-constrained cooperative planning of multi-agent systems.
In the first part of the thesis, given a team of rigid bodies, we propose model-free and decentralized control protocols such that a desired distance and orientationbased formation between neighboring agents is achieved. Inter-agent collisions are guaranteed to be avoided by the proposed control scheme. Furthermore, the connectivity between agents that are initially connected is preserved. The transient and steady state responses are solely determined by certain designer-specified performance functions.
In the second part of the thesis, the problem of robust navigation of a multi-agent system to predefined states of the workspace while using only local information is addressed, under certain distance and control input constraints. The agents are modeled by nonlinear continuous-time dynamics with additive and bounded disturbances. In order to address this problem, decentralized tube-based nonlinear Model Predictive Control protocols are proposed. In particular, the feedback control law contains a portion that is calculated offline and a portion which is the outcome of an online optimal control problem.
In the third part of the thesis, a team of agents operating in a bounded workspace is considered. Each agent is assigned with high-level tasks given in Metric Interval Temporal Logic. First, by providing novel decentralized abstraction design techniques, the motion of each agent is captured through a weighted transition system. Then, we propose decentralized control methodologies and high-level algorithms that guarantee the satisfaction of the desired tasks of each agent. The proposed approach can handle couplings as well as transient constraints of each agent in a novel way.