Obstacle avoidance by profit sharing using self-organizing map-based probabilistic associative memory

Abstract

In this paper, we realize an action learning of obstacle avoidance by Profit Sharing using Self-Organizing Map-based Probabilistic Associative Memory (SOMPAM). In this method, patterns corresponding to the pairs of observation and action are memorized to the SOMPAM, and the brief degree is set to value of the rule. In this research robot learns with the aim of acquiring an action rule that can reach the goal point from the start point with as few steps as possible while avoiding collision with the obstacle. We use the reduced image of the image taken with the small camera mounted on the robot as observation. In the simulation environment reproducing the experimental environment, we confirmed that the learning converged to a state where it can reach the goal while avoiding obstacles with the minimum steps. Moreover, even in the real environment, it was confirmed that the robot can reach the goal while avoiding obstacles.