Upkie wheeled biped robot

Upkie is a fully open source self-balancing wheeled biped robot. It has wheels for balancing and legs to negotiate uneven terrains. Upkies are designed to be buildable at home using only tools and components ordered online, like mjbots actuators. You can develop in Python or C++, on Linux or macOS, then deploy your software to the robot's Raspberry Pi.

This repository contains all the materials needed to build and animate an Upkie:

• Hardware:
  • Bill of materials
  • Build instructions
  • Project log
• Software:
  • Getting started
  • Installation
  • Example
  • Documentation
• Going further:
  • More agents
  • More examples

Questions are welcome in the Chat and Discussions forum.

Getting started

You can play with a balancing Upkie right away from your command line:

git clone https://github.com/upkie/upkie.git
cd upkie
./try_pid_balancer.sh

Click on the robot in the simulator window to apply external forces. If you have a gaming controller connected to your computer, steer its joysticks to move the robot around 🎮

Installation

Conda

conda install -c conda-forge upkie

PyPI

pip install upkie

Example

You can develop your own agent using the Gymnasium environments distributed in upkie.envs. For example, here is a simple proportional-feedback balancer:

import gymnasium as gym
import upkie.envs

upkie.envs.register()

with gym.make("UpkieGroundVelocity-v3", frequency=200.0) as env:
    observation, _ = env.reset()
    action = 0.0 * env.action_space.sample()
    for step in range(1_000_000):
        observation, reward, terminated, truncated, _ = env.step(action)
        if terminated or truncated:
            observation, _ = env.reset()
        pitch = observation[0]
        action[0] = 10.0 * pitch

To run this agent, you will first need to start a simulation:

./start_simulation.sh

Then execute the Python code above. To run the agent on the robot, do the same but running the pi3hat spine instead of the simulation.

Agents

The demo agent simply balances in place by PID control. There are more advanced Upkie agents distributed in their own repositories. Check out the ones you are interested in:

• MPC balancer: balance in place using model predictive control.
• Pink balancer: an extended PID balancer than can crouch and lift its legs.
• PPO balancer: balance in place with a policy trained by reinforcement learning.

Head over to the new_agent template to create your own.

Frequently Asked Questions

• Is the Python API fast enough for real-time control?
• Can we step a Gym environment faster than real time?
• How to pair a Bluetooth controller to the Raspberry Pi?

Citation

If you built an Upkie or use parts of this project in your works, please cite it as follows:

@software{upkie,
  title = {{Upkie open source wheeled biped robot}},
  author = {Caron, St\'{e}phane and Perrin-Gilbert, Nicolas and Ledoux, Viviane and G\"{o}kbakan, \"{Umit} Bora and Raverdy, Pierre-Guillaume and Raffin, Antonin},
  license = {Apache-2.0},
  url = {https://github.com/upkie/upkie},
  version = {3.4.0},
  year = {2024}
}