Event excitation for event-driven control and optimization of multi-agent systems

TL;DR

This paper introduces a new metric for event-driven control in multi-agent systems that ensures event excitation even in environments with discrete points of interest, enhancing the effectiveness of gradient-based optimization.

Contribution

It proposes a novel potential field metric that guarantees event excitation in stochastic hybrid systems, improving event-driven control in multi-agent environments.

Findings

The new metric ensures non-zero gradient values without events.

Application to multi-agent data collection demonstrates effectiveness.

Numerical examples validate the approach.

Abstract

We consider event-driven methods in a general framework for the control and optimization of multi-agent systems, viewing them as stochastic hybrid systems. Such systems often have feasible realizations in which the events needed to excite an on-line event-driven controller cannot occur, rendering the use of such controllers ineffective. We show that this commonly happens in environments which contain discrete points of interest which the agents must visit. To address this problem in event-driven gradient-based optimization problems, we propose a new metric for the objective function which creates a potential field guaranteeing that gradient values are non-zero when no events are present and which results in eventual event excitation. We apply this approach to the class of cooperative multi-agent data collection problems using the event-driven Infinitesimal Perturbation Analysis (IPA) methodology and include numerical examples illustrating its effectiveness.