
Asimov 1
Asimov 1 is Menlo Research's open-source bipedal humanoid, and the interesting part is not what it does but what it gives away. Mechanical CAD, electrical CAD, the bill of materials, circuit schematics and a MuJoCo simulation model are all public, under CERN-OHL-S-2.0 for the hardware and GPL-2.0 for the software, so you can source every part yourself and build one, or buy the Here Be Dragons Edition kit for a USD 499 deposit against a target price of around USD 15,000 and spend more than 100 hours assembling it. What arrives is a 1.2 metre, 35 kilogram biped with 25 actuated degrees of freedom plus two passive toe joints, and the mechanical choices show what it is for: an RSU revolute-spherical-universal ankle with parallel actuation for more natural ground reaction forces, arms of 5 degrees of freedom each ending at wrist yaw, a single waist rotation and a 2 degree-of-freedom neck. It produces up to 120 newton metres of peak torque and lifts 5 kg per arm rated, 15 kg at peak. Two computers run it, a Raspberry Pi 5 for multimedia and networking and a Radxa CM5 for real-time motion control, over five 1 Mbps CAN buses and one 500 kbps bus, and because the simulation model ships with the hardware a policy trained in simulation runs on the same robot. Perception is deliberately modest and aimed at locomotion and manipulation research: a 2MP camera, a quad-microphone array, a 6 degree-of-freedom IMU, joint state feedback and a speaker. If you do not own one yet, the browser-based Digital Asimov twin runs your code against the same model. Menlo sells the platform as hardware for builders, engineers and researchers rather than as a worker, and the company behind it is small, self-funded and employee-owned, with the follow-up Asimov 2 still in progress.
Research on locomotion and manipulation policies against real hardware; raw data collection from a running robot; live demonstrations driven by joystick, code or an agent; education and self-sourcing build projects; and simulation-to-real transfer work, since a policy trained in simulation runs unchanged on the robot and a browser-based twin allows code to be tried without hardware.
Get weekly updates on humanoid robot funding, new companies, and product launches.