王兴兴上市后首次公开演讲：机器人真正爆发要等两个「80%时刻」丨WRC 2026

TL;DR - Unitree founder Wang Xingxing argues that robots will reach mass adoption only when they can complete roughly 80% of tasks across 80% of unfamiliar environments. The company is pursuing “self-evolving” physical AI that connects research, code generation, simulation, real-robot testing, and feedback into an automated development loop.

• Generalization, rather than hardware capability, remains the main bottleneck: small environmental or object changes can sharply reduce task success rates.
• Unitree is developing multimodal, end-to-end systems that generate robot actions from spoken instructions, though current execution has several seconds of latency and limited motion fluency.
• Wang identifies model-to-physical-world alignment—especially final-centimeter or millimeter errors—as a key obstacle to reliable manipulation.
• The proposed self-evolution pipeline uses coding agents to create control software, validates it in simulation and on physical robots, then feeds AI and human evaluations back into subsequent iterations.