In November 2016, Adam Conway and Will Roscoe launched the Donkey Car project. The idea that drove both was to give interested hobbyists a possibility to enter the world of self-driving cars in model format. The most important reasons for me to use the Donkey Car framework are the large international community, the free availability of the software, the Donkey Car Simulator and the fact that standard modeling technology is used to build the physical robot car. The use of standard model cars reduces the costs significantly.

Originally, the Donkey Car framework was developed for the Raspberry Pi, which is the best selling single board computer and with its huge community helped lay the foundation for the success of the Donkey Cars project. With the availability of the NVIDIA Jetson Nano 2019 which has a powerful GPU architecture and its similar ARM processing unit as used by the Raspberry Pi, the Donkey Car community immediately recognized the benefits and additionally extended the framework with support for the Jetson Nano. One of the biggest advantages of the Jetson Nano compared to the Raspberry Pi is that neural network training is possible directly on the Jetson Nano thanks to its GPU architecture. With the built-in GPU units, the Jetson Nano can perform parallel calculations that allow it to train neural networks in a high-performance manner. The Raspberry Pi 4 with 4 GB RAM but without GPU support has no chance against the Jetson Nano when it comes to executing and training neural networks. The Donkey Car Framework written in Python and its good inline documentation additionally facilitate the introduction to the project of autonomously driving robot cars and to the topic of artificial intelligence. At this point, the simulator for the Donkey Car should also be mentioned. This allows you to get started with the Donkey Car project in a very easy and really inexpensive way. This is possible because you can install the Donkey Car Simulator on existing hardware such as a laptop or PC without further ado.

My conclusion in sum is that the Donkey Car framework in its combination of simulator solution and physical model car is very practice-oriented, easy to understand and therefore ideal for getting started with artificial intelligence.

Donkey Car E-Book

No gray theory but a practical introduction to:

  • Artificial Intelligence and Deep-Learning
  • Simulation and generation of synthetic data
  • Training of neural networks
  • Testing the neural network in the real world
  • Step by step instructions for building a robot car

E-Book – Download:

English: 20210627-Donkey-Car-E-Book-v-0-1-0-en-Ingmar-Stapel

Deutsch: 20210627-Donkey-Car-E-Book-v-0-1-0-de-Ingmar-Stapel

Installationsanleitung – Windows

The Video below describes how to install the Donkey Car Framework and Simulator on a Windows 10 PC

Installationsanleitung – Ubuntu

The Video below describes how to install the Donkey Car Framework and Simulator on a Ubuntu 20.x PC

Donkey Car – Simulator race

The following video shows four neural networks driving in parallel against each other in a Donkey Car Unity environment in a fully autonomous car race. The neural networks were trained in such a way that they find their way back to the race track by driving backwards when they hit an obstacle.

Donkey Car – RC model

In this video you can see the real Donkey Car driving fully autonomously. The construction of exactly this Donkey Car is described in the free e-book here on my website.

(Visited 1,097 times, 1 visits today)