Microduck

Microduck is an open-source tiny biped robot from Pollen Robotics; in the README's words, "A tiny biped robot that moves using reinforcement learning policies." This repository holds the software that runs on the robot, using an ONNX neural network policy to control 15 servos 50 times a second; training lives in a separate repository, microduck_rl, which uses MuJoCo and PPO. What sets it apart is that you can run a simulation on your computer with scripts/duck-sim before you have a robot.

Who it's for:robotics and reinforcement learning developers who have the hardware or are happy to simulate first. Readers who don't write code won't get much from it.

Before you start: the physical robot is bought separately, and the repository doesn't say what it costs. The runtime includes system services for motor control, WiFi and identity setup, Bluetooth, camera video over WebRTC and signed automatic updates, all with high privileges. A physical robot really moves and can bump into people or things, so it isn't an ordinary web toy. The code is under the Apache-2.0 licence; check the licences of the training repository and hardware files separately rather than assuming the same terms.

Based on the project's public documentation. Not tested by AI Inspiration World.