Unitree CEO Wang Xingxing Calls Model Generalization Biggest Robotics Bottleneck
At the 2026 APEC CEO Summit, Unitree founder and CEO Wang Xingxing stated that model generalization is the most challenging bottleneck for robots today, noting that robots should be able to work immediately in any household without requiring scene-specific retraining. This highlights the core gap between digital AI and physical robotics, emphasizing that true commercial scaling in homes and industries depends on overcoming software adaptability rather than just hardware design. Wang also noted that a technological breakthrough threshold is approaching rapidly, but scaling production from tens of thousands to millions of units and satisfying unprecedented computing power demands will present massive secondary challenges.
## BACKGROUND
Generalization in robotics refers to a machine's ability to adapt and perform effectively in unseen environments and tasks without requiring extensive retraining. Embodied AI seeks to combine artificial intelligence algorithms with physical robotic bodies, allowing them to interact intelligently with the real world.