embodied AI

A Chinese startup developing 'robot brains' — AI systems that let robots perceive and act in the physical world — says it could reach a ChatGPT-style breakthrough as soon as next year. The claim suggests the kind of leap seen in language models may soon arrive in embodied AI, a field where Chinese firms are racing US rivals. Observers are watching whether the company can deliver on such an ambitious timeline.

A new benchmark called Roboharm asks whether frontier AI robot policies refuse unsafe instructions, such as commands that could cause physical harm when executed by embodied systems. The project, hosted on RoboCurve, is drawing attention among AI safety researchers and robotics practitioners who are debating how well current vision-language-action models handle hazardous requests.