We build humanoid robots in Penang, Malaysia, and publish how they are made: the CAD, the actuators and the software that teaches them to walk. Our robots are modular, low-cost and function-first, built on a standard set of actuators in three sizes (Macro, Micro and Nano), so the same parts serve many joints.
Try Luma-1 in your browser: steer the trained robot live and see how it tested.
Our first humanoid: about 1 m tall, with 19 motors and shock-absorbing ankles. It learned to stand and walk in simulation, in Isla Studio, before it takes a step on the floor. Its CAD ships with Studio under the MIT licence.
Macro actuator High torque. A 36:1 cycloidal drive with low-side inline current-sensing FOC control. |
Micro actuator Medium torque. A 7.385²:1 two-stage planetary gearbox, driven by ESP32 SimpleFOC with an AS5047P magnetic encoder. |
| Repository | What it does |
|---|---|
| isla-studio | Teach a legged robot to stand and walk on your own computer. Import a URDF or MJCF, train it with reinforcement learning in MuJoCo, test it in full physics and run it on the real robot over CAN. Comes with Luma-1 and trained examples. |
| Live demo | Robots trained in Studio, walking live in your browser at isla-demo.vercel.app. |
Everything we publish is MIT-licensed. Issues and pull requests are welcome on every public repository.