Unitree CEO Wang Xingxing calls robot model generalization the hardest problem today
Wang Xingxing said at the Shenzhen APEC Business Leaders Summit that generalization in robot models is the hardest problem right now, and expressed hope that robots can work in homes immediately without retraining.
Unitree founder Wang Xingxing said at the Shenzhen APEC Business Leaders Summit that generalization in robot models is the hardest problem facing the industry today. He expressed hope that robots can go into ordinary households and start working immediately, without retraining for each new task. Responding to questions about when humanoid robots will take on real work in home settings, he said that may become possible in the not too distant future.
Generalization remains the bottleneck mainly because mainstream robot training still depends heavily on data and manual demonstrations collected for specific tasks, leaving transfer to new scenes, new objects and different hardware platforms limited. Recent industry responses include scaling cross-embodiment data, unified modelling through vision-language-action models, and continuous adaptation through reinforcement learning on the deployment side.
Wang made the remarks as funding in embodied AI heated up globally, with European humanoid robot firm Genesis AI closing a 500 million dollar Series A at a 3.5 billion dollar valuation on the same day, while domestic Chinese robot makers pushed ahead with factory pilot deployments. He has repeatedly argued that commercialization of humanoid robots should progress from structured settings such as factories and warehouses toward unstructured environments such as homes.