Reinforcement learning of path-finding behaviour by a mobile robot

We describe how a simple autonomous mobile robot can learn to navigate towards a goal while avoiding obstacles. A neural network determines the actions of the robot in response to the inputs from an array of infrared sensors. A reinforcement learning algorithm adjusts the weights of the neural network until the appropriate "action mapping" from sensor input to action output is found. Learning takes place in real time in the robot. The learning method is generic and therefore suitable for any robot with similar sensor and effectors.