Abstract
Most of existing robot learning methods assume that the environment where a robot works does not change, therefore, a robot has to learn from scratch if it encounters new environments. This paper proposes a method which adapts a robot to environmental changes by efficently transferring a learned policy in the previous environments into a new one and effectively modifying it to cope with the environmental changes. The resultant policy (a part of state transition map) does not seem optimal with respect to each individual environment, but may absorb the differences between multiple environments. We apply the method to a mobile robot navigation problem of which task is to reach the target avoiding obstacles based on uninterpreted sonar and visual information. Experimental results show the validity of the method.
