Singapore's Asimov is opening up its v1 humanoid robot, giving builders the mechanical files, simulation tools, and parts list needed to experiment outside a closed hardware stack.
Asimov has released v1, a full-scale open-source humanoid robot built for developers who want to build, modify, and train their own machines. The project includes CAD files for structural parts, a MuJoCo simulation model for locomotion work, and a complete bill of materials. The robot stands 1.2 meters tall, weighs 35 kilograms, and has 25 actuated degrees of freedom plus two passive toe joints. Its arms are rated to lift about 15 kilograms in flexion and about 18 kilograms laterally.
That matters because humanoid robotics is still dominated by expensive, tightly controlled systems. Boston Dynamics, Tesla Optimus, Unitree, Figure, and other players have pushed the category forward, but most builders still cannot easily inspect the design, replace parts, or adapt the hardware to their own research. Asimov is taking the opposite approach. The v1 kit is being offered for a $499 refundable preorder deposit, with a target price of $15,000, while builders can also source parts themselves from the published bill of materials.
According to Asimov's own manual, the robot uses 7075 aluminum and MJF PA12 nylon in its structure, with a Raspberry Pi 5 handling media and networking, a Radxa CM5 for motion control, and CAN bus connections for the actuators. The company says the kit will arrive unassembled, with manuals, build videos, and Discord support for people willing to do the mechanical and electrical work themselves. That makes v1 less like a polished consumer robot and more like a serious reference platform for labs, startups, and advanced builders.
The project comes from Menlo Research, Asimov's parent company, after the team ran into the same problem many robotics developers face: closed hardware slows everything down. A Unitree G1 knee failure reportedly left the team waiting about two months for a replacement part, pushing Menlo toward a design that could be inspected, repaired, and rebuilt with standard components. Asimov first released its leg design in January, including six degrees of freedom per leg, an RSU ankle architecture, and passive toe joints, before expanding the work into the full humanoid v1 release under the CERN-OHL-S-2.0 license.
V1's most useful idea is modularity. Universal motor mounts are designed to let users swap legs, arms, and torso components without rebuilding the entire robot. The RSU ankle uses a parallel two-motor drive to share pitch and roll torque, while the simulation setup helps developers map policies from MuJoCo into the physical machine. The assembly process is still demanding, but it gives teams a documented path through the mechanical and electronic work instead of asking them to reverse engineer a black box.
Menlo is also using v1 as the hardware base for its broader software stack. That includes an agent platform, Uranus for behavior validation in simulation, and Cyclotron for training locomotion and manipulation skills. In practice, the robot is meant to become a testbed where builders can train policies, validate behavior before deployment, and collect data from cameras, audio, IMUs, and motor states. The GitHub release turns that from a company-only workflow into something outside developers can fork, test, and improve.
The pitch is simple: make humanoid robotics feel more like Raspberry Pi or Arduino and less like a sealed industrial product. Standard parts, 3D-printable components, and a public design lower the cost of learning by doing. That does not make v1 cheap or easy. A $15,000 target kit is still serious money, and the self-sourced bill of materials can run higher at low volume. But compared with closed humanoid systems that can cost well into six figures or remain unavailable to outside developers, the barrier is meaningfully lower.
Open Wins the Ecosystem Race
The open-source strategy is also a bet on speed. Proprietary robotics companies control their hardware, software, repairs, and developer access because reliability matters. Asimov is betting that a wider community can surface fixes, test new configurations, and build useful integrations faster than one closed team can do alone.
That timing is important. Humanoid robotics has pulled in billions of dollars as advances in AI, simulation, perception, and dexterity make the category look more practical. Warehouses, logistics sites, and manufacturing floors are the obvious early targets, but deployment is still constrained by hardware cost, safety, maintenance, and the shortage of people who can build and test useful behaviors. An open reference design gives smaller teams a way into that loop.
Asimov's model still carries risk. Open hardware invites variation, and variation can create quality-control problems. Support can become expensive if builders underestimate the difficulty of assembling and tuning a humanoid robot. Safety also becomes harder when more people can modify the machine. The CERN license encourages collaboration, but it does not remove the engineering burden that comes with moving 35 kilograms of actuated hardware through the real world.
Still, the playbook is familiar. Arduino did not win because it was the most powerful electronics platform. Raspberry Pi did not succeed because it replaced every computer. They won because they gave developers a practical place to start, then let communities form around shared parts, documentation, and examples.
For founders, that is the real lesson. Platforms rarely become important just because the first version is impressive. They become important when outside builders can extend them, repair them, and make the next use case cheaper for everyone else. Asimov wants v1 to play that role for humanoid robotics.
The next signal will not be the launch announcement. It will be what the community does with the files. Watch the GitHub forks, early walking policies, replacement-part designs, and first third-party builds. If builders can assemble v1, improve it, and share useful changes, Asimov becomes more than another humanoid maker. It becomes an ecosystem coordinator in a market still waiting for its standard development platform.
Also read: Joby Aviation's JFK-Manhattan test flight puts air taxis seven minutes from reality • Ted Lieu's AI bill cracks down on deepfakes and shields whistleblowers • Microsoft drops OpenAI exclusivity and ends revenue share in landmark deal restructure
