Learning to Play Air Hockey with Model-Based Deep Reinforcement Learning
This addresses the Robot Air Hockey Challenge 2023, providing an incremental improvement in robotic control for dynamic competitive games.
The researchers tackled the problem of training a robotic manipulator to autonomously play air hockey using model-based deep reinforcement learning with sparse rewards and self-play, achieving stable learning and better performance with longer imagination horizons.
In the context of addressing the Robot Air Hockey Challenge 2023, we investigate the applicability of model-based deep reinforcement learning to acquire a policy capable of autonomously playing air hockey. Our agents learn solely from sparse rewards while incorporating self-play to iteratively refine their behaviour over time. The robotic manipulator is interfaced using continuous high-level actions for position-based control in the Cartesian plane while having partial observability of the environment with stochastic transitions. We demonstrate that agents are prone to overfitting when trained solely against a single playstyle, highlighting the importance of self-play for generalization to novel strategies of unseen opponents. Furthermore, the impact of the imagination horizon is explored in the competitive setting of the highly dynamic game of air hockey, with longer horizons resulting in more stable learning and better overall performance.