Abstract
In recent years, robots are used as partner in the field of service and healthcare. Therefore, robots are needed to decide actions autonomously to communicate smoothly with human. For action decision, observing of antagonist reactions, learning and replacing antagonist reactions as self-performed reactions are needed. First in observation part, robots observe antagonist reactions for self-performed actions. Then, robots learn form observation of antagonist reactions and replace antagonist reactions for self-performed actions by self-performed reactions for antagonist actions. We established an action decision algorithm for robots based on imitation of antagonist actions. We expressed learning of robots by changing learning coefficient. We expressed action decisions of robots by selecting learning coefficient. We confirmed the efficacy of the algorithm by the result of sensory evaluation.
