TikTok owner builds AI 'brain' for household robots
What's the story
ByteDance, the Chinese tech giant behind TikTok, has unveiled a robotics system powered by a large-scale AI model, showcasing its ambition in the field of embodied intelligence. The innovative technology, pairs the new GR‑3 AI model with ByteMini, a lab‑stage bimanual mobile robot, to perform household tasks like hanging a shirt or clearing a table.
Tech details
GR-3 can understand and execute natural language instructions
The GR-3 is a large-scale vision-language-action model that enables robots to understand and execute natural language instructions. It can perform general tasks on unseen items, in new environments, or with abstract concepts relating to size and spatial relationships. During a demo, the robot accurately inserted a hanger into a shirt and hung it on a rack, even though its training only included long‑sleeved garments. ByteDance describes GR‑3 as a scalable "brain" for robots—bridging vision, language and action in one architecture.
Versatility
GR-3 system can also perform complex tasks
The GR-3 system isn't just limited to simple tasks. It can also pick up an individual item from a few pieces and place it on a designated spot. A robot powered by this AI model can identify an object not just by its name but also by its size or spatial relationships. This makes the technology extremely versatile and adaptable to different situations. ByteDance aims to build general-purpose robots that can operate in real-world environments using this innovative model.