The Paris Olympics are in full swing, and the intense competition in table tennis has spectators holding their breath. Even Google couldn't resist stepping in, unveiling its first robot agent, Agent, which has reached human competitive levels.

Table tennis, a sport that requires years of training to master, demands players to possess complex basic skills and strategic gameplay abilities. This sets it apart from purely strategic games like chess and Go.

Thus, table tennis has become a valuable benchmark for measuring robot capabilities, testing their abilities in high-speed movement, real-time precision, strategic decision-making, system design, and direct competition against human opponents.

To train this robot, Google collected a vast amount of table tennis state data, including positions, speeds, and spins. Through learning this data, the robot mastered skills such as topspin forehand, backhand aiming, and serve reception.

It first underwent training in a simulated environment, accurately mimicking the physical characteristics of table tennis matches. Subsequently, the robot was deployed into the real world, collecting performance data through matches with humans to improve its skills, forming a continuous feedback loop.

Additionally, Google designed the system to adapt to different opponents by tracking their behaviors and playing styles. This allows the robot to try different skills, monitor its success rate, and adjust strategies in real-time.

To evaluate the robot's skill level, Google arranged competitive matches against 29 table tennis players of varying skill levels. Results showed that the robot won 45% of the matches and 46% of the games against all opponents. Notably, it achieved a clean sweep against beginners, while losing all games against advanced and advanced+ players. This indicates that the robot has reached an intermediate human level in endurance matches.

Trial participants highly praised the experience of playing against the robot, finding it both fun and engaging regardless of their skill level. They expressed a willingness to play again. While advanced players could exploit the robot's strategic weaknesses, they still enjoyed the matches and saw its potential as a more dynamic practice partner.

Despite certain limitations such as reaction speed, camera sensing capabilities, spin handling, and paddle rubber, the robot's breakthrough performance in table tennis is still remarkable. The introduction of Google's robot Agent not only brings new vitality to the sport of table tennis but also opens up new possibilities for the development of robotics.

Project Homepage: https://sites.google.com/view/competitive-robot-table-tennis/home