Towards Adversarial Training for Mobile Robots

Abstract

This paper reports some preliminary work on learning on a physical robot. In particular, we report on an experiment to learn how to strike a ball to hit a target on the ground. We compare learning based just on previous trials with the robot with learning based on those trials plus additional data learnt using a generative adversarial network (GAN). We find that the additional data generated by the GAN improves the performance of the robot.