This paper presents a learning mechanism allowing a multi-agent system to cooperate to achieve a gathering task efficiently in unknown and changing environments. The multi-agent system is a team of autonomous behaviour-based agents with limited communication capabilities. Cooperation is based on the acquisition of signaling behaviours and on the specialization of the agents into different types. Every agent has the same collection of built-in-reactive behaviours. Some of the built-in behaviours are fixed, while others can be modified through reinforcement learning. The reinforcement signal is delayed until the completion of a trial and assesses the collective performance of the team. Each agent uses this common signal to learn what individual behaviours are more suitable for the team. Simulation results, and the corresponding statisticla analysis, show that the multi-agent system always achieves near-to-optimal performances.

MURCIANO Antonio; 

1996-09-09

MIT Press

JRC13774

https://publications.jrc.ec.europa.eu/repository/handle/JRC13774,