Unitree founder Wang Xingxing: embodied AI tipping point could come in one to two years

At CIIF 2026, Unitree's founder tied the timeline to next-generation AI models and said the company is testing a ‘physical AI self-evolution’ loop.

Wang Xingxing, founder of Unitree Robotics, said at the opening-ceremony roundtable of the 26th China International Industry Fair on October 12 that embodied AI's tipping point could arrive quickly — within one to two years at the fastest. He expects a next-generation foundation model release next year could break through the field's current bottlenecks.

He pointed to the pace of AI progress, with major updates landing every month or two, and said Unitree's tests of the latest models on its robots have performed well. Looking back on more than a decade of work, he said his biggest mistake was not believing in AI strongly enough — a conviction he says has only grown.

Internally, Unitree is testing what he called a ‘physical AI self-evolution’ system: a self-reinforcing loop in which AI handles code writing, model training, deployment, testing, and feedback. Humans still sit in the evaluation loop — judging whether motions match expectations, tasks succeed, and outcomes are genuinely what people want — but he expects human involvement to taper off as experience accumulates.

On commercial reality, Wang said Unitree has run pilot deployments since 2024 across scenarios including automotive factories, but full-scale rollout will take more time. Industrial settings and commercial To-B services will mature faster than the home, he said.

The forecast is his most optimistic yet. In August 2025 he put the ‘ChatGPT moment’ for humanoids at one to two years in the best case (three to five at the slowest); this past August he widened the range to two-to-three years at best and five-to-ten at slowest. The roundtable remarks move the fast end of the range back to one to two years.