Self-learning neural control of a mobile robot

Reinforcement learning is a promising paradigm for the training of intelligent controllers. The learning capabilities of a neural network based controller architecture are shown by its application to control a mobile robot in an unknown environment. Based on multi-sensor information provided by four infrared sensors, the controller has to learn to avoid collisions, receiving only a nal training signal of success or failure. The article further shows, that simulation can be used to avoid long real world training eeort.