ByteDance Seed's robotics research team, developer of the GR-3 vision-language-action model and the ByteMini bimanual mobile robot.
The Seed Robotics Team is the robotics research group within ByteDance Seed, the AI research organization of ByteDance. The team works on generalist robot policies, and it is listed as the author of the GR-3 technical report published on the ByteDance Seed website and on arXiv (2507.15493) on 21 July 2025.
GR-3. GR-3 is a large-scale vision-language-action (VLA) model intended as a generalist robot policy. According to the team, it generalizes to novel objects, environments and instructions involving abstract concepts, and it can be fine-tuned with minimal human trajectory data for fast, low-cost adaptation to new settings. It handles long-horizon and dexterous tasks, including bimanual manipulation and tasks that require mobile movement. The training recipe combines co-training on web-scale vision-language data, fine-tuning on human trajectory data collected with VR devices, and imitation learning on robot trajectory data. In the team's real-world experiments GR-3 outperformed the pi-0 baseline across a range of challenging tasks; media coverage showed the system folding clothes and clearing tables.
ByteMini. Alongside GR-3 the team introduced ByteMini, a bimanual mobile robot that carries out many different tasks when paired with GR-3. ByteMini is presented as a research platform; no commercial availability or hardware specifications have been published.
Primary type & automation activities this supplier delivers:
Large-scale vision-language-action model for generalist, long-horizon bimanual and mobile robot manipulation.
Bimanual mobile research robot built to run GR-3 across a wide range of household-style tasks.
Contact Seed Robotics Team, ByteDance Seed
WEBSITE
https://seed.bytedance.comHEADQUARTERS
China
Company Facts
Founded
-
Primary Role
Research Institute
Company Size
-
Primary Region
Asia-Pacific
Annual Sales
-
Funding Stage
Subsidiary / Acquired
Funding Total
-
Keywords