Reinforcement Learning for Trajectory Following
Gerhard Neumann, Michael Pfeiffer and Helmut Hauser
Graz University of Technology, Graz, Austria.
In this tech report we investigate the use of reinforcement learning techniques for point to point and trajectory following movements of a simulated center of mass of a humanoid robot. The task is to reach a given point as fast as possible without violating the given ZMP constraints (which assure that the robot is not falling). We tested several different reinforcement learning algorithms with different settings.