Gasgoo Munich-AgiBot announced on June 15 that its full-size bipedal humanoid robot, the Yuanzheng A3, has successfully mastered autonomous table tennis. This marks the first time a full-size bipedal humanoid has completed a match with fully autonomous decision-making throughout the entire process.
Image Credit: Screenshot from AgiBot's official video
Table tennis is unforgiving: balls travel at speeds exceeding 5 meters per second, with spin, trajectory, and landing points shifting in a split second. Yet the Yuanzheng A3 handled it all without remote control, scripts, or human intervention. It executed a closed-loop control system—processing visual perception, predicting trajectories, planning whole-body movements, and striking the ball with precision.
This breakthrough stems from a collaboration between Agibot and a team led by Zhang Shanghang at Peking University. They leveraged SpikePingpong—the world's first table tennis motion control algorithm for humanoid robots—alongside a 20kHz high-frequency spike camera developed by Professor Huang Tiejun. The result? Visual response speeds ten times faster than conventional solutions, enabling millimeter-level prediction of where the paddle meets the ball and significantly boosting control precision.
Powered by this new algorithm architecture, the robot operates free of human remote control. It engages in fully autonomous play, returning shots continuously while switching fluidly between offense and defense.
Demonstrating human-like perception and motor skills in high-speed dynamic interactions suggests these robots are ready for real-world deployment. As its capabilities evolve, AgiBot is moving quickly beyond basic mobility toward understanding, collaboration, and execution—laying the groundwork for the era of general-purpose robotics.
