Actuation constraints and artificial physics control

Abstract

Swarm systems for multiagent control rely on natural models of behavior. Such models both predict simulated natural behavior and provide control instructions to the underlying agents. These two roles can differ when, for example, controlling nonholonomic robots incapable of executing some control suggestions from the system. We consider a simple physicomimetics system and examine the effects of actuation constraint on that system in terms of its ability to stabilize in regular formations, as well as the impact of such constraints on learning control parameters. We find that in the cases we considered, Physicomimetics is surprisingly robust to certain types of actuation constraint. © 2008 Springer-Verlag Berlin Heidelberg.