Researchers in China have developed EyeVLA, a robotic eyeball that mimics human vision by rotating and zooming to capture fine details something fixed RGB-D cameras struggle with. Powered by a vision-language-action model trained with reinforcement learning, EyeVLA can follow instructions, shift its viewpoint, and focus on specific targets. In indoor tests, the system delivered clearer, more accurate visual information without relying on expensive sensors. The team says EyeVLA could enhance embodied AI in tasks like infrastructure inspection, warehouse monitoring, environmental observation, and household robotics.
