Multiactor approach and hexapod robot learning

Abstract

This paper presents a multiactor approach of the Q-learning used to teach a hexapod robot to control its trajectory. So, each actor participating to the same global task performs its own learning process taking into account or not the other agents. As any actor "leg of hexapod robot" cannot achieve its movements without interacting with others, co-ordination may be set up. This "co-ordination with actors" approach is applied to solve the problems of displacement, trajectory and posture control of a hexapod robot in its environment. The efficiency of the approach is validated through simulation results.