Enactic open-sources OpenArm, a 7DOF humanoid arm for physical AI research

An open arm priced for labs

Enactic has published OpenArm, a fully open-source humanoid arm with seven degrees of freedom, and the project's GitHub repository has drawn front-page attention on Hacker News. According to the project's documentation, the arm is built for physical AI research and for deployment in contact-rich environments, combining high backdrivability and compliance for safe interaction with people, with payload capabilities the team describes as practical for real-world applications.

The arm has human-scale proportions, and a complete bimanual system is priced at $6,500. It can be bought assembled or as a DIY build, sourced from manufacturers the project lists as verified and certified. The intended workloads span teleoperation, imitation learning, simulation and real-world data collection in contact-rich tasks.

Nine repositories, two licences

The release is split across nine repositories. The hardware side, openarm_hardware, carries the complete CAD data, including STL files, STEP files and Fusion 360 assemblies, under the CERN-OHL-S-2.0 licence, the strongly reciprocal variant of the CERN Open Hardware Licence, which requires derivative hardware designs to stay open under the same terms.

Everything else ships under Apache-2.0: robot description files with URDF and xacro for simulation, a CAN control library for low-level motor communication, ROS 2 integration packages, teleoperation packages supporting unilateral and bilateral control, an Isaac Lab simulation environment with training tasks, MuJoCo specification files and assets, a dataset repository covering the data format, recording tools and a Python API, and Dora dataflow nodes for data collection, inference and teleoperation.

Taken together, the stack covers the full loop of a modern robot-learning workflow: collecting demonstrations through teleoperation, training in simulation, deploying through ROS 2 or Dora, and logging episodes into a common dataset format, all on one platform whose design anyone can inspect and modify. The licence split is notable: hardware derivatives must remain open, while the permissive software licence lets labs fold the code into larger internal or commercial stacks.

Standardised cells for comparable results

Alongside the arm, the project is promoting OpenArm Cell, a standardised physical environment that fixes the backdrop, lighting and camera positions. The stated goal is that research performed on OpenArm can be reproduced anywhere in the world under consistent evaluation conditions, letting results from different labs be compared directly.

That is an unusual commitment for hardware research. Manipulation results are notoriously sensitive to camera angle, lighting and workspace layout, one reason machine-learning-style benchmarks have been slow to carry over to physical robotics. By pinning those variables down, OpenArm Cell aims to make progress claims checkable across institutions instead of each group reporting numbers from a slightly different bench.

Why it matters

Robot learning has moved fast in simulation, but progress on real hardware is bottlenecked by access. Proprietary research arms are expensive and closed, and results gathered on one platform often cannot be reproduced on another. OpenArm addresses both problems at once: open CAD and control software make the platform inspectable and modifiable, while the $6,500 bimanual price puts a human-scale, compliant system within reach of university groups and independent researchers.

The breadth of the release matters as much as the price. Because simulation models, teleoperation tooling, dataset tooling and deployment packages all target the same arm, a lab can work end to end on a single stack rather than integrating components from several vendors. The team says it is actively looking for contributors, research partners and company collaborators, and it runs a documentation site and a Discord community for coordination. How large an ecosystem forms around the platform will decide whether OpenArm becomes a common benchmark for physical AI research or another well-documented design that stays on its creators' benches.