Unitree's Wang Xingxing Calls Security Robots a Coming Growth Area, Citing Overseas Demand

Unitree founder and chairman Wang Xingxing told a business conference in Hangzhou on November 6 that he sees security as a promising next application for robots, saying overseas demand for intelligent security robots is strong, Jiemian News reported. The remarks were made at the eighth Zhejiang entrepreneurs' conference and are an outlook, not a product launch.

During the talk, Wang showed video of Unitree robots in security and emergency scenarios, including restraining a simulated attacker, stopping a vehicle trying to force its way through, and emergency rescue. Jiemian described these as demonstration clips; Unitree has not announced a dedicated security product, customers or field deployments in connection with the remarks, and the clips do not establish how the robots perform in real incidents.

Much of the speech emphasized that robots can do practical work rather than only perform. Wang said Unitree has pursued factory applications since 2020, working with several automakers and auto-parts suppliers as well as in its own plant. He acknowledged that humanoid robots worldwide have not yet entered large-scale trials or widespread use and remain less efficient than conventional industrial automation.

In fixed settings such as factories, he said, robots trained for a specific task can approach a 100% success rate, but homes remain the biggest obstacle because layouts vary so widely. His stated goal is a robot that can work autonomously in any environment without prior training.

Wang predicted that next year or the year after, robots may complete around 80% of tasks in completely unfamiliar environments from voice instructions alone, which he called the moment embodied AI reaches its equivalent of a breakthrough consumer chatbot. The main obstacle, he said, is that model outputs and physical robot actions are not yet precisely matched, leaving errors of a few millimeters.

He also argued that AI progress now comes less from new algorithms than from automated loops of data collection, scoring, training and refinement, making computing power and chip capacity decisive, and said the same logic could apply to humanoid robots once large-scale self-training works. He compared today's robots with personal computers more than 20 years ago, when many people saw them mainly as gaming machines because software was immature.

SEE ALSO: Unitree Releases UniUni Swarm Control Software, Free for Its Users