AgiBot A3 Plays Autonomous Table Tennis

Bipedal humanoid table tennis is a powerful benchmark for embodied AI that highlights the critical need for ultra-low-latency perception in dynamic environments.

* The integration of a 20kHz spike camera demonstrates that sensor speed is just as vital as compute power for real-time physical reaction.

* Co-developing the SpikePingPong algorithm with Peking University shows that standard reinforcement learning models benefit heavily from specialized, hardware-aligned visual algorithms.

* Achieving millimeter-level trajectory planning on a full-size humanoid marks a significant step forward in whole-body coordination and physical interaction.