Strategy and Skill Learning for Physics-based Table Tennis Animation

Abstract

Figure 1: On the left, two physically-simulated agents engage in a competitive match, controlled by our strategy and skill controllers. The center panels show the five learned skills: Forehand Drive, Push and Smash, Backhand Drive and Push, showcasing the skill controller's ability to execute a diverse set of skills. On the right, a human interacts with the simulated agent through VR.