Google's AI company DeepMind developed an AI robotic system that can play table tennis at an intermediate amateur level, winning around 45% of matches and 46% of individual games against human players of varying skill levels. It excelled against beginners, won 55% against intermediate players, but lost to advanced players.
The robot combines an industrial robotic arm with custom AI software, with high-speed cameras to track the ball and respond effectively to human players' moves. It can execute various shots, read ball tilts, and respond to different ball speeds.
The robotic system was trained on a comprehensive dataset of table tennis ball states and practiced skills like forehand topspin and backhand targeting in a simulated environment. Its training involved reinforcement learning, imitation learning, and learning from real-world ball trajectories and human-vs-human gameplay data.
The robot adapted to different playing styles by tracking opponents' behaviors and adjusting its strategy through repeated training cycles involving reinforcement learning with real-world data.
Table tennis has been a standard for robotic research due to its demands on skills and planning. The achievement is viewed as a step towards achieving consistent human-level performance in table tennis with robots.
Researchers propose investigating advanced control algorithms and hardware optimizations to improve the robot's performance, particularly in handling fast balls, and making its play more unpredictable by allowing it to learn from human opponents' strategies.
The research represents progress towards creating robots that can perform tasks skillfully and safely in real environments. Playing against the AI robot was generally fun and engaging for most human players.
The robot arm is mounted on linear gantries for sideways movement and holds a 3D-printed paddle. The AI system has low-level skill controllers and a high-level strategic decision-maker, allowing it to execute table tennis techniques and adapt its strategy.
Table tennis is chosen by many robotics companies, including OMRON's "FORPHEUS" robot table tennis tutor, to train their systems due to its requirements.
The robot could potentially be used to help athletes train and create unexpected plays. Table tennis was chosen as a sport due to the complex physical and cognitive elements involved.
Sources: Technology Review, Live Science, TechRadar, Digit, The Next Web, Times of India, The Valley Post, Brytfmonline, Ex Bulletin, NY Breaking, The Times Hub, Singularity Hub, Techeblog, Faharas, Designboom, Tech Xplore, Cybernews, Interesting Engineering, NewsBytes.
This article was written in collaboration with Generative AI news company Alchemiq.
