ACTION-BASED SENSOR SPACE SEGMENTATION FOR SOCCER ROBOT LEARNING

Robot learning, such as reinforcement learning, generally needs a well -defined state space in order to converge. However, building such a st ate space is one of the main issue of robot learning because of the in terdependence between state and action spaces, which resembles the wel l-known ''chicken and egg'' problem. This article proposes a method of action-based state space construction for vision-based mobile robots. Basic ideas to cope with the interdependence are that we define a sta te as a cluster of input vectors from which the robot can research the goal state or the state already obtained by a sequence of one kind of action primitive regardless of its length, and that this sequence is defined as one action. To realize these ideas, we need many data (expe riences) of the robot and we must cluster the input vectors as hyper e llipsoids so that the whole state space is segmented into a state tran sition map in terms of action from which the optimal action sequence i s obtained. To show the validity of the method, we apply it to a socce r robot that tries to shoot a ball into a goal. The simulation and rea l experiments are shown.