Section: New Results

Predictive Control for multi-agent (multi-vehicle) systems

Participants : Sorin Olaru [correspondent] , Ionela Prodan [LCIS] , Minh Tri Nguyen [L2S] , Cristina Stoica [L2S] , Silviu Niculescu [L2S] , Fernando Fontes [FEUP U. Porto] , Joao Sousa [FEUP U. Porto] , Fernando Lobo Pereira [FEUP U. Porto] , Alexandra Grancharova [U. Sofia, Bulgaria] .

We continued a mature line of research on the tracking problems for multi-agent systems. In [75] we presented a series of developments on predictive control for path following via a priori generated trajectory for autonomous aerial vehicles. The strategy partitions itself into offline and runtime procedures with the assumed goal of moving the computationally expensive part into the offline phase and of leaving only tracking decisions to the runtime. First, it will be recalled that differential flatness represents a well-suited tool for generating feasible reference trajectory. Next, an optimization-based control problem which minimizes the tracking error for the nonholonomic system is formulated and further enhanced via path following mechanisms. Finally, possible changes of the selection of sampling times along the path and their impact on the predictive control formulation will be discussed in detail.

On a relatively different frameowrk, in [71] we investigate multiple agents evolving in the same environment with the objective of preservation of a predefined formation. This formation aims to reinforce the safety of the global system and further lighten the supervision task. One of the major issues for this objective is the task assignment problem, which can be formulated in terms of an optimization problem by employing set-theoretic methods. In real time the agents will be steered into the defined formation via task (re)allocation and classical feedback mechanisms. The task assignment calculation is often performed in an offline design stage, without considering the possible variation of the number of agents in the global system. These changes (i.e., including/excluding an agent from a formation) can be regarded as a typical fault, due to some serious damages on the components or due to the operator decision. In this context, our work proposes a new algorithm for the dynamical task assignment formulation of multi-agent systems in view of real-time optimization by including fault detection and isolation capabilities. This algorithm allows to detect whether there is a fault in the global multi-agent system, to isolate the faulty agent and to integrate a recovered/healthy agent. The proposed methods will be illustrated by means of a numerical example with connections to multi-vehicle systems.