Collective Iterative Learning Control: Exploiting Diversity in Multi-Agent Systems for Reference Tracking Tasks

  • 2021-04-15 17:36:00
  • Michael Meindl, Fabio Molinari, Dustin Lehmann, Thomas Seel
  • 1

Abstract

This paper considers a group of autonomous agents learning to track the samegiven reference trajectory in a possibly small number of trials. We propose anovel collective learning control method (namely, CILC) that combines IterativeLearning Control (ILC) with a collective input update strategy. We deriveconditions for desirable convergence properties of such systems. We show thatthe proposed method allows the collective to combine the advantages of theagents' individual learning strategies and thereby overcomes trade-offs andlimitations of single-agent ILC. This benefit is leveraged by designing aheterogeneous collective, i.e., a different learning law is assigned to eachagent. All theoretical results are confirmed in simulations and experimentswith two-wheeled-inverted-pendulums robots (TWIPRs) that jointly learn toperform a desired maneuver.

 

Quick Read (beta)

loading the full paper ...