Abstract
Humans can rapidly adapt to unexpectedly changing environment and achieve various tasks. Specifically, the number of trials for adaptation is small when current situation is similar to the memorized ones, suggesting evaluation of similarity would be a key component to boost learning speed. Here, we develop a model for learning obstacle avoidance motions which can store various input-output mappings with respect to a situation, and can recall them depending on similarity of the situation. Simulation results demonstrate that a vehicle can learn how to overcome obstacles, even though position and property of which is not given in advance.
