Motion Control of a Robot by Means of Q-Learning Using the Example of Locomotion

The aim is to program a robot who learns a movement using Q-Learning. A neural network is used to process the data. To check whether the robot has learned successfully, four requirements were made of the robot. Various experiments were carried out with the robot, in which the parameters of the learning algorithm or the neural network were varied. The results of the experiments show that the robot fulfills every requirement, except in the experiments where the neural network diverges and the robot does not learn any strategy to reach the goal.