Abstract
In multi-robot systems, role assignment is important for appropriate cooperation. From the viewpoint of robustness, we have proposed a new concept of multi-robot cooperation termed "autonomous specialization". This concept means that robots adaptively develop roles and assign them by themselves. As a methodology for autonomous specialization, we have been developing a reinforcement learning (RL) technique, termed BRL. In our previous research, autonomous specialization in multi-robot systems in observed. The concept is extended as that of human-robot cooperation. In this paper, we conduct physical experiments using two arm-type robots with BRL, the task of which is lifting up an object by cooperating with a human partner. Then the autonomous specialization process is analyzed.
