Open Hardware & RISC-V

Seeed reBot Arm B601-RS targets open VLA research with NVIDIA Isaac GR00T integration

Key Takeaway

Seeed's $2,999 open-source 6-DOF robotic arm bridges industrial precision and hobby affordability for vision-language-action research, with validated NVIDIA Isaac GR00T and Hugging Face LeRobot support.

Summary

Seeed Studio has released the reBot Arm B601-RS, a desktop 6-DOF robotic arm designed specifically for Physical AI and vision-language-action (VLA) research. The arm offers 0.1 mm repeat positioning accuracy, 2.5 kg rated payload, and 754.7 mm reach, using 48V Robstride Dynamics quasi-direct-drive motors with CAN bus communication. It is explicitly validated for NVIDIA Isaac GR00T fine-tuning and Hugging Face LeRobot integration, with full hardware drawings, BOM, Python SDK, ROS 2 Humble drivers, and USDA simulation models published on GitHub.

The B601-RS was developed as reference hardware for a co-created NVIDIA course on sim-to-real VLA pipelines, launched at the 2026 World Robot Conference. It supports the complete Physical AI workflow: teleoperation and dataset collection, Isaac Sim digital twin emulation, Cosmos Transfer data augmentation, GR00T 1.7 policy training, and Jetson edge deployment. Beyond the NVIDIA ecosystem, it also supports ROS 1/2, Pinocchio, MuJoCo, and custom Python tooling.

Hardware architecture and open-source stack

  • Joint actuation: 48V Robstride Dynamics quasi-direct-drive motors provide stronger torque and faster dynamic response compared to 24V-class alternatives, with CAN bus communication via USB2CAN adapter and MotorBridge platform for calibration.
  • Precision specifications: 0.1 mm repeat positioning accuracy and low-backlash design aim to reduce demonstration noise that compromises VLA training data quality on hobby-grade servo arms.
  • Open-source release: Hardware drawings (sheet metal and 3D-printed parts), full BOM, assembly videos, 48V power supply design, Python SDK, ROS 2 Humble drivers, and USDA simulation models are available in the Seeed-Projects/reBot-DevArm GitHub repository.
  • Ecosystem integration: Official Hugging Face LeRobot integration and validated NVIDIA Isaac GR00T compatibility, including leader-follower teleoperation rigs for synchronized RGB-D stream and joint trajectory recording.
  • Simulation and deployment: USDA-format digital twin assets for NVIDIA Isaac Sim, with documented workflows for Cosmos Transfer domain randomization and GR00T 1.7 dual-system (vision-language reasoning + diffusion action head) deployment to NVIDIA Jetson.

Why it matters

The reBot Arm B601-RS addresses a persistent gap in open robotics by offering industrial-grade precision at a desktop price point with fully open hardware and software stacks, lowering barriers for academic labs and independent researchers to replicate and extend VLA research. Its validated GR00T integration and published sim-to-real pipeline provide a concrete reference implementation that could reduce fragmentation in Physical AI tooling and accelerate reproducible robotics research across the Linux and ROS ecosystems.

Original reference: www.seeedstudio.com →