WHRG Table Tennis Debuts on One 1.73-Meter Platform
Table tennis is new at the World Humanoid Robot Games. Global Times reported that all 12 teams competed on the same 1.73-meter humanoid platform, so the scoreboard is about software, not a taller torso or a fancier actuator pack.
Tang Minqin, who chairs the table tennis organizing committee, told the paper a robot has about 0.3 seconds to see the ball, predict, decide, and swing. No remote control, no scripts, no canned rallies. Teams had less than three months after the event was added in June. By day 3 the field had moved from groups into knockouts.
Same body, different brain
That 1.73-meter lock is the interesting rule. Last year’s games let hardware differences shout. Here the organizers stripped that out. If your robot misses, it is the perception loop and the swing planner, not a secret motor.
The Atlantic still from August 22 shows the scene: cowboy hat versus ball cap, orange ball in the air, crowd phones up. The near robot wears an AgiBot / Shanghai Jiao Tong University bib. I am not treating that jersey as a gold medal. Global Times had not named a champion when it published on August 24.
Tang’s other line is the one factories will quote. Training a robot to track a fast ball, decide, and coordinate a swing, he said, can transfer to industrial work. “If these training methods can teach a robot to play table tennis, they can also potentially be transferred to industrial scenarios.”
Why this is harder than a sprint
A 100-meter heat is a gait and a stop. Ping pong is a closed loop at ball speed. AP’s broader Games coverage still leads on sprints and records. The table-tennis rulebook is doing different work: same machine, autonomy required, a clock measured in tenths of a second.
AgiBot previously posted a full-size Yuanzheng A3 autonomous table-tennis match in June, with Gasgoo covering a collaboration with Zhang Shanghang’s Peking University group and a 20 kHz spike camera. That was a different robot and a company demo. The Games event is the standardized 1.73-meter class Tang described.
A Human’s Take
Locking the hardware is the best rule this meet has written. It stops the “our legs are 10 cm longer” argument and leaves you with a camera, a predictor, and a 0.3-second budget. I will care about the winner when someone posts rally length and unforced-error counts, not a single highlight smash.