ByteDance Unveils AI Brain for Robots: A Leap Towards Everyday Automation

The TikTok-owning company ByteDance has unveiled a system that acts as the «brain» for robots, enabling them to handle everyday tasks such as hanging up clothes or clearing a table.

The GR-3 is an advanced language model that integrates vision and action, allowing bots to follow commands expressed in natural language and perform a range of tasks involving unfamiliar objects. These robots are capable of adapting to new environments or working with abstract concepts related to size and spatial relationships.

A video shared on the company’s website showcases how the lab’s two-armed ByteMini robot can insert a hanger into a shirt and place it on a rack.

In a separate technical report, the team noted that the bot can manage short-sleeve garments, despite the training data exclusively consisting of long-sleeve items.

Thanks to GR-3, the robot can execute commands to choose a specific item from several options and place it in a designated location.

The system can identify objects not just by their names, but also by size (for example, «large plate») or spatial cues (such as «to the left»). It can fully accomplish the task of «clearing the dining table» with a single command.

For training the model, ByteDance employed a multi-faceted approach that includes various components.

«We hope that GR-3 will pave the way for the development of versatile robots that can assist people in their daily lives,» the team stated.

Previously, in January, the startup Perplexity AI announced its intention to acquire the U.S. division of TikTok. The company sent ByteDance a proposal to merge Perplexity, TikTok U.S., and new capital partners into a single legal entity.