Should I Trust my Teammates? An experiment in Heuristic Multiagent Reinforcement Learning
Reinaldo A. C. Bianchi, Ramón López de Mántaras · DIGITAL.CSIC (Spanish National Research Council (CSIC)) · 2009
Trust and reputation are concepts that have been traditionally studied in domains such as electronic markets, e-commerce, game theory and bibliometrics, among others. More recently, researchers started to investigate the benefits of using these concepts in multi-robot domains: when one robot has to decide if it should cooperate with another one to accomplish a task, should the trust in the other be taken into account? This paper proposes the use of a trust model to define when one agent can take an action that depends on other agents of his team. To implement this idea, a Heuristic Multiagent Reinforcement Learning algorithm is modified to take into account the trust in the other agents, before selecting an action that depends on them. Simulations were made in a robot soccer domain, which extends a very well known one proposed by Littman by expanding its size, the number of agents and by using heterogeneous agents. Based on the results it is possible to show the performance of the learning algorithm can be improved even when using very simple trust models. 1