Asimov 1: an open source humanoid robot design

What the repository actually contains

Asimov is an open-source humanoid robot that you can build, train and customize, and this repository holds version 1 of it. The README is precise about what is inside. The repository contains the mechanical CAD, electrical CAD, simulation model and onboard software to build, simulate and customize Asimov 1. The top level of the tree matches that list, with a `mechanical/` directory, an `electrical/` directory, a `sim-model/` directory, a `scripts/` directory, an `assets/` directory holding the render of the robot, and two licence files, `HARDWARE-LICENSE.txt` and `SOFTWARE-LICENSE.txt`, which is a split that tells you the project treats the machine and the code as separately licensed artefacts. There is also a `.gitmodules` file, which means the simulation model or another component is pulled in as a submodule rather than vendored. The primary language recorded for the repository is Python, the last push was on 2026-10-06, and the homepage is the documentation site for version 1.

The physical specification

Asimov 1 is a 1.2 m, 35 kg biped with 25 actuated degrees of freedom, and the specification table breaks that down in a way that tells you what the machine can actually do. The legs are 6 DOF each, so 12 of the 25. The arms are 5 DOF each, described as shoulder pitch, roll and yaw, elbow and wrist yaw, so 10 more. The torso carries 1 DOF of waist yaw, a 10 W 4 ohm speaker and a 6 DOF IMU. The head has 2 DOF of neck, neck yaw and neck pitch, a stereo microphone array and a 2MP monocular camera. The CAN bus is 5 channels at 1Mbps and 1 at 500kbps. Onboard compute is a Raspberry Pi 5 for media and network plus a Radxa CM5 for motion control, and the structural materials are 7075 aluminium and MJF PA12 nylon. The rated and peak load table gives a squat rated load of 5 kg, a bicep curl and front raise at 5 kg rated and 15 kg peak per arm, and a lateral raise at 6 kg rated and 18 kg peak per arm.

Two licences, and why the split is normal here

The two badges at the top of the README make the licensing position explicit. The hardware is under CERN-OHL-S-2.0, with the file named HARDWARE-LICENSE.txt, and the software is under GPL-2.0 with SOFTWARE-LICENSE.txt. That pairing is the sensible arrangement for a robot, and the reason is that the CERN Open Hardware Licence is designed for exactly this kind of artefact. It is a copyleft licence for hardware, so a derivative design has to be released under the same terms, but it does not impose the software copyleft obligations on the machine itself in the way a GPL on the whole repository would. A repository under a single GPL would have been awkward, since the CAD files and the simulation model are not programs and applying a software licence to them creates ambiguity about what you may do with a bracket. Two files and two badges remove the ambiguity, and anyone designing a derivative should read HARDWARE-LICENSE.txt rather than assume the software licence covers the drawings.

Two ways to build it, and what each includes

The README lays out two options. The first is the DIY kit, everything needed to build Asimov 1 unassembled, shipping now for $20,000. The inclusion table is explicit about what is in the box and what is not. Hardware includes all BOM components unassembled, the power supply and cabling, and spare parts, while tools and hands are not included. Compute includes the RPi edge board, motion control board, network board and power distribution board, but not the 4G and 5G modules. Sensors include the monocular camera, IMUs, microphone, speaker and motor joint states, but not lidar or a 360 camera. Docs include the quick start guide, the manual and the DIY build videos. The second option is self-sourcing, where you pull the BOM from the documentation and fabricate everything yourself, and the README says to start with the repo-local fabrication manifest for the CAD-derived part inventory, then cross-reference the BOM and assembly manual for procurement details, sourcing and fabrication notes.

The fabrication manifest, and the script that checks it

The fabrication manifest is the piece that makes self-sourcing tractable, and it is worth understanding what it is. It is described as the CAD-derived fabrication manifest for the part inventory, and it lives in the repository as `mechanical/FABRICATION_MANIFEST.csv`. The point of a derived manifest is that it is generated from the CAD rather than maintained by hand, which means the bill of materials and the drawings cannot drift apart the way two hand-maintained documents would. The script that produces it is in the repository and can be run locally to check the manifest:

python3 scripts/generate_fabrication_manifest.py --check

The `--check` flag is the useful one for a builder, because it verifies the committed manifest against the CAD without rewriting it. That is a small piece of tooling that says something about how the project is run. If the check passes, the inventory in the CSV matches what the models actually contain, and your procurement list is derived from the design rather than from a document that was correct at some point in the past.

The roadmap, and what is not done yet

The roadmap table is short and uses status marks. Four items are complete. Mechanical CAD is done across 7 subassemblies. The MuJoCo simulation model is done. The electrical wiring harness is done. And electrical schematics and PCB files are done. Two are marked as upcoming. Asimov Edge is next, which from the naming is the software layer that runs on the machine, and a locomotion policy is next, which is the trained behaviour that makes it walk. The distinction between the two is worth pausing on, because it explains what the completed parts are worth on their own. The CAD, the harness and the simulation model let you build a machine and test control code against a model of it, but a humanoid that does not walk is a very expensive shelf. So the repository as published is the mechanical and electrical foundation plus a simulation environment, and the behaviour is the part still to come. The presence of the simulation model is what makes the gap workable, since a locomotion policy can be trained against it before the hardware exists.

The project around the repository

The repository is one artefact of a larger operation, and the links in the header describe the rest. There is the project website, a manual, a DIY kit page, a 3D model of the machine, a Discord server, an X account and a forum. The manual is linked separately from the documentation site, and the BOM and assembly manual are hosted there, while the 3D model is served as a standalone HTML file, which suggests it is meant to be opened in a browser rather than downloaded into a CAD tool. The ways to get involved are split by audience. Build questions go to the forum or to a GitHub issue for bugs and contributions, and the issue link in the README points at a different repository, the asimovinc/asimov-v1 one. Deploying Asimov has a direct mailto link, and supply chain partners, meaning manufacturers of actuators, structural components or electronics, are invited to make contact. The licence situation and the price of the kit together say what kind of project this is: a commercial product with its hardware design opened up, not a hobby project that happens to be free.

Editorial conclusion

Adopt the asimov-1 design if you are building a humanoid and want a complete mechanical and electrical package plus a simulation model to iterate against before hardware exists. Skip it if you want a robot you can assemble from a kit on a normal budget, because the DIY kit is stated at $20,000 and the self-source route assumes you can fabricate parts. Before committing, read both licence files, since the hardware is under CERN-OHL-S-2.0 and the software under GPL-2.0, and check the fabrication manifest locally with `python3 scripts/generate_fabrication_manifest.py --check` before you start procurement.

Frequently asked questions

Official sources

1. Issues
2. License: CERN-OHL-S-2.0
3. menloresearch/asimov-1 on GitHub
4. Project website
5. README