Abstract
This paper shows how to include a crude but efficient communication component into reactive robotic multi-agent systems. The basis of the work relies on the concept of self-satisfaction of an agent composed of a selfish tendency and an altruistic one. A signal, representing the variation of an agent’s “interactive satisfaction”, is emitted in its neighborhood. It measures the expected efficiency of cooperation. The information propagates from agent to agent within a certain range. It is shown how the information is converted into a vector able to alter instantaneously the agent’s motion when added to the classical vectors which generate the selfish motion: move-to-goal and avoid-obstacle. Then the paper presents the architecture for implementing such mechanisms to achieve robust adaptive cooperation. The results of simulation experiments are presented which demonstrate the efficiency of the generalized dynamic potential field including communication of satisfaction. In particular, the method allows the agents to escape from a deadlock situation, and makes the collective behavior converge towards a steady flow of traffic.