Outplaying elite table tennis players with an autonomous robot Artificial intelligence systems have increasingly demonstrated capabilities to match or exceed human performance in computer games, yet physical sports like table tennis remain a significant challenge due to their demand for rapid, precise, and adversarial interactions. A new autonomous system, named Ace, has been developed to compete with elite human players, marking a breakthrough in real-world robotics. Ace combines advanced perception and control technologies to achieve competitive performance in high-speed, high-spin table tennis matches. Table tennis requires split-second reactions, with ball speeds often exceeding 20 meters per second and shot intervals under 0.5 seconds. The spin of the ball, which can reach 1,000 radians per second, further complicates gameplay by altering trajectories and making shots harder to return. Previous robots have struggled with these dynamics, often relying on simplified setups such as limited court coverage, fixed ball launchers, or ignoring spin entirely. These limitations have hindered their ability to replicate the complexity of human play. Ace addresses these challenges through a novel approach. It features a high-speed perception system utilizing event-based vision sensors, which enable real-time tracking of the ball and racket. The control system is based on model-free reinforcement learning, allowing the robot to adapt dynamically to unpredictable human opponents. Additionally, Ace employs state-of-the-art high-speed robot hardware, ensuring precise execution of shots. Evaluated in matches against elite and professional players under official competition rules, Ace achieved several victories and demonstrated consistent performance in generating high-speed, high-spin shots.#ai #ace #table_tennis #reinforcement_learning #robotics
