Open-Source Robotics in Practice: Lessons from Upkie Wheeled Bipeds

In this talk, we will discuss the open hardware and open-source software available today for creating open-source robots, taking the Upkie wheeled bipeds as our working example:

• Project page: https://hackaday.io/project/185729-upkie-wheeled-biped-robots
• Robot software: https://github.com/upkie/upkie/
• Robot hardware: https://github.com/upkie/upkie_parts

On the software side, we will go through the core robotics libraries used by the robots: Gymnasium, moteus, and PyBullet, as well as libraries for developing robot behaviors, such as Stable-Baselines3 for reinforcement learning and qpmpc for model predictive control. We will compare model-based and AI-driven approaches to implement new behaviors on Upkies, discussing what has worked and not worked so far, as well as future plans for the project like integrating vision for perceptive locomotion.

On the hardware side, we will describe how Upkies use mjbots actuators and PCB cards, which are fully open source (down to motor-driver firmware and KiCad electronics files!). Upkies can be built from off-the-shelf components for around $3,000, making experimentation with bipedal robots more accessible to hobbyists and educators. We will finally conclude by releasing version 2 of the robot hardware with an on-stage demonstration.