autorope/donkeycar

What it solves

Donkeycar provides a minimalist, modular Python library for building and teaching small-scale self-driving robots. It lowers the barrier to entry for hobbyists and students to experiment with autonomous driving, computer vision, and neural networks without requiring deep prior expertise.

How it works

The library uses a pipeline of software "parts" that run in a loop on the vehicle. These parts handle specific tasks such as capturing images from cameras or lidar, reading GPS data, receiving steering/throttle inputs from controllers, and controlling the drivetrain motors. Users can choose from pre-built templates or create custom parts to modify the car's behavior. For autonomy, it supports three types of autopilots: deep-learning (via TensorFlow, TensorFlow Lite, and PyTorch), GPS-based, and computer vision-based.

Who it’s for

It is designed for hobbyists, high school and university students, and researchers who want to learn about autonomous driving and robotics through hands-on experimentation.

Highlights

• Modular Architecture: Easily swap or add software parts to the vehicle's loop.
• Multiple Autopilot Options: Supports deep learning, computer vision, and GPS-based navigation.
• Hardware Agnostic: Compatible with various cameras, lidar, and drivetrains (e.g., RC cars, differential drive).
• Simulation Support: Includes a simulator for testing self-driving logic before building physical hardware.
• Rich Interface: Features a graphical user interface and WebUI for remote control and configuration.