Reinforcement learning for VR table tennis

This paper investigates the use of Reinforcement Learning (RL) to train an Artificial Intelligence (AI) humanoid opponent to play a Virtual Reality (VR) table tennis game. A self-play RL algorithm is implemented to train an AI opponent through competing against itself in Unity environment. The simulation environment replicates key physics of table tennis, and the agent controls racket movements to hit the ball. Experimental result indicates that the agent progressively enhances its table tennis skills according to rising ELO rating and optimization of aggressive gameplay strategies. This research provides a valuable framework for utilizing RL to overcome limitations of scripted AI opponents in physics-based sports games. RL opens new possibilities for human-like AI that can provide dynamic and adaptive experiences in VR games.