Kinetix AI and HKU MMLab publish SMASH humanoid table-tennis research
SMASH uses onboard egocentric vision and learned whole-body control for humanoid table-tennis strikes on a Unitree G1.
Evidence notes
- The system perceives the ball through onboard egocentric cameras rather than an external motion-capture setup.
- Its learned controller coordinates full-body strokes including smashes and low crouching returns.
- The paper reports consecutive physical-robot rallies and explicitly identifies Unitree G1 as the experimental platform.
Company context
Kinetix AI is a Shenzhen-based embodied AI and humanoid robotics company developing ultra-human-like robots. Its KAI platform centers on whole-body control, spatial intelligence, embodied foundation models, tactile sensing, and whole-body data acquisition.