Unitree chief says embodied AI software still trails robot hardware

At Beijing’s World Robot Conference, Unitree chief executive Wang Xingxing said software—not hardware—is the principal constraint and described a future target in which a robot could enter an unfamiliar home and complete roughly 80% of requested tasks from natural-language instructions. reports say the comments after Unitree’s volatile post-IPO trading.

Why it matters: The admission is a useful counterweight to humanoid demonstrations: robust perception, planning, manipulation and recovery in unfamiliar environments remain unsolved.

Business/technical/strategic impact: Treat the 80% figure as an aspiration, not a demonstrated benchmark. Procurement should require task-level success, safety cases, intervention rates, maintenance costs and performance outside staged environments.