Meet Booster T2: An Open Humanoid Platform for Embodied AI Development

ROS ROS General
1

We’re excited to introduce Booster T2, our flagship humanoid development platform for robotics researchers and developers.

T2 is designed to support work across:

  • Whole-body locomotion and control

  • Mobile manipulation and dexterous-hand research

  • Perception, navigation, and human–robot interaction

  • Teleoperation and real-world data collection

  • Reinforcement learning, imitation learning, and VLA development

  • Onboard AI applications and external system integration

The platform provides ROS 2 compatibility, high- and low-level SDK access, onboard NVIDIA computing, and support for custom sensors and applications.

Product: Booster T2 | Flagship Embodied Development Platform
Documentation: https://docs.booster.tech/
GitHub: BoosterRobotics · GitHub

We’re interested in connecting with researchers, developers, and robotics teams working on embodied AI and real-world humanoid applications. If you would like technical information, pricing, or a T2 demonstration, please contact me at chenshaoyun@booster.tech.

— Viki Chen
Booster Robotics

1 Like
ROS News for the Week of August 24th, 2026
2

Nice platform. One question I ask everyone shipping an open humanoid, and I ask it because I maintain the tooling and I still cannot answer it for my own work.

Does the T2 stack ship any adversarial evaluation of the policy layer, or is the safety story entirely at the controller and hardware layer?

Reason for asking: whole-body policies have a failure mode that a joint-limit check does not see. A perturbed instruction can produce a trajectory that is inside every mechanical limit and still wrong. We have a humanoid suite for exactly this, and honestly it is the suite where our own action-envelope defence gets no credit, which is a result I would rather share than hide.

If the answer is “not yet”, that is the normal answer, and I am not scoring anyone. Happy to compare notes.