Multiagent coordination utilising Q-learning

Abstract

The problem 'Coordination' among multiple agents is an active and open-ended problem of robotics. The coordination problem, considered in this paper, has been divided into three categories: (1) Spatial coordination; (2) Temporal coordination and (3) Spatio-temporal coordination. The principles of 'Situation Calculus' have been extended to model the spatial coordination problem of robot. The concept of temporal coordination has been introduced with Situated Automata and Q-learning. The collective learning behaviour of a multiagent system has been improved by the principles of Q-learning. The spatio-temporal coordination, that deals with the coordination problem involving both space and time has been modelled using the timed Petri nets. The special emphasis has been given to the behavioural model of eye and hand coordination of a mobile robot. One typical application of the multiagent coordination in the soccer playing robot has been proposed at the end of this paper.