NVIDIA and Archon's PingpongDuo Teaches Two Humanoid Robots to Rally Together
A hierarchical system lets two humanoid robots alternate shots in table tennis doubles, clearing space in 2.4 seconds without colliding.
- PingpongDuo is billed as the first humanoid table tennis doubles system with two robots alternating strikes.
- A learned risk filter vetoes positioning plans likely to cause teammate collisions in shared workspace.
- Separate footwork and striking skills are trained from human motion capture, including real doubles footwork.
- A racket-arm residual policy corrects swings after fast motion using a calibrated sim.
- Beats SOTA baseline by 13.2% return rate; 2.15 m clearance in 2.4 s on hardware.
- Project page with 3D demo: archon.tech/research/PingpongDuo.
Humanoid robots have played table tennis solo for several years, but doubles requires two machines to coordinate on the same side of the table. Researchers from the Shanghai Innovation Institute, Archon Robotics, the University of Hong Kong, and NVIDIA have presented PingpongDuo, which they describe as the first humanoid doubles system. The robots can sustain more than 10 consecutive strikes in strict alternation, with demonstration footage showing an 18-return rally.
Table-tennis doubles combines rapid whole-body control with a shared-space scheduling problem. Teammates must hit every other ball while repositioning in a confined area, often with less than two seconds before the next return. Overlapping paths can cause a collision, while excessive retreat can leave a robot too far away to swing accurately. The controller must decide where both robots should stand, which robot has priority, and whether each can reach its next position in time.
Three layers keep the robots apart
The paper divides coordination and control across three layers:
- Joint team planning. A planner selects positioning targets for both robots. A learned risk estimator rejects target combinations likely to cause a collision and reserves space around the robot currently due to strike.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
