Menlo Research uwalnia Asimov v1: CAD, MuJoCo i licencja CERN-OHL-S

Menlo Research has published the full source for the Asimov v1 humanoid: CAD files, MuJoCo simulation models, mechanical and electronic documentation and control software. Everything has been released to a public GitHub repository under CERN-OHL-S-2.0 (hardware) and GPL-2.0 (code). The decision means any research team or independent engineer can build, modify and commercially use the platform, provided they preserve copyleft — that is, release their own modifications under the same license.

Key facts

Platform: Asimov v1 humanoid, around 1.2 m tall, 25 degrees of freedom.
Licenses: CERN-OHL-S-2.0 for hardware and CAD files, GPL-2.0 for software — both copyleft.
Repository: github.com/asimovinc/asimov-v1.
Simulation: ready-to-run MuJoCo models, enabling reinforcement learning training before assembly.
DIY kit cost: around $15,000 according to project documentation.

What was actually open-sourced

The package includes CAD files for the body, limbs and drive assemblies, a full bill of materials (BOM), control electronics schematics and onboard software. The repository also contains MuJoCo models that are ready to run, so researchers can train control policies in simulation before assembling a physical robot.

According to the documentation, Asimov v1 has 25 degrees of freedom, including arms, hands and legs. Around 1.2 m tall and a few dozen kilograms, the platform sits in the "small humanoids" segment alongside Unitree G1 and Booster T1 — but with one critical difference: the full mechanical design is open and modifiable.

Why CERN-OHL-S-2.0 matters

CERN-OHL-S-2.0 is a strong copyleft license for open hardware — the GPL of the hardware world. If anyone uses the project commercially (e.g. builds a product on top of the CAD files), they must publish their modifications under the same license. This guarantees the ecosystem that future versions stay open.

It is meaningful because, until now, "open source" in humanoid robotics typically meant publishing individual components or AI models, not full mechanical documentation. Menlo Research combines both into a single complete bundle.

Where it sits in the humanoid market

In 2026 the humanoid market splits into two clear streams. On one side — large commercial players (Tesla Optimus, Figure AI, Apptronik, 1X) with closed hardware and proprietary models. On the other — platforms from the Chinese market (Unitree, Booster Robotics) sold as products but with very limited access to design documentation. Menlo Research introduces a third option: a fully open-hardware platform that you can build yourself.

The direction here lines up with earlier academic efforts (Berkeley HumanPlus, Stanford Aloha) — but Asimov v1 is the first project in which CAD, software and full MuJoCo simulation models are open at the same time, under copyleft licenses.

Why it matters

Open-sourcing Asimov v1 lowers the entry barrier for humanoid research from "top university lab only" to "small research teams, startups and technical education". For reinforcement learning research in humanoid robotics, the ability to verify results in the physical world — not just in simulation — is critical. Asimov v1 provides that ground truth at a fraction of the cost of commercial platforms. The second effect is standardization: if the community builds a set of models and benchmarks around Asimov v1, the platform could become the reference hardware for academic publications — the role Franka Panda and Universal Robots UR5 played in the world of robotic arms. The third effect is long-term: copyleft enforcement ensures the ecosystem will not be "closed" by the first large commercial user.

What's next

First independent reproductions — teams that build Asimov v1 from CAD and publish results.
External RL benchmarks based on the Asimov v1 MuJoCo model.
Response from commercial humanoid companies — whether any vendor will go open-source at a comparable scale.