Abstract
This paper describes an effect of knowledge transfer rate for transfer learning of heterogeneous multiple robots. In recent years, various researches for a real-world application of multiple autonomous robots system have been discussed. In a dynamic environment such as RoboCup soccer, a Multi-Agent Robot Learning System (MARL) using reinforcement learning is useful to obtain adaptive behaviors autonomously. The transfer learning shows an effectiveness of reuse of knowledge to several issues of MARL such as a long learning time. However, in the transfer learning of heterogeneous multiple robots, the effect of the transfer rate, which means a ratio of knowledge transfer between robots, is not discussed. In this paper, we confirmed feature of change of transfer rate in the heterogeneous multiple robots' environment by conducting a computer simulation.
