Dynamic movement primitives and reinforcement learning for adapting a learned skill

摘要

Traditionally robots have been preprogrammed to execute specific tasks. This approach works well in industrial settings where robots have to execute highly accurate movements, such as when welding. However, preprogramming a robot is also expensive, error prone and time consuming due to the fact that every features of the task has to be considered. In some cases, where a robot has to execute complex tasks such as playing the ball-in-a-cup game, preprogramming it might even be impossible due to unknown features of the task. With all this in mind, this thesis examines the possibility of combining a modern learning framework, known as Learning from Demonstrations (LfD), to first teach a robot how to play the ball-in-a-cup game by demonstrating the movement for the robot, and then have the robot to improve this skill by itself with subsequent Reinforcement Learning (RL). The skill the robot has to learn is demonstrated with kinesthetic teaching, modelled as a dynamic movement primitive, and subsequently improved with the RL algorithm Policy Learning by Weighted Exploration with the Returns. Experiments performed on the industrial robot KUKA LWR4+ showed that robots are capable of successfully learning a complex skill such as playing the ball-in-a-cup game.