DIY Robocar projects come in all shapes and sizes, and to suit all budgets. If you’ve got the parts lying about (such as a radio control car, Raspberry Pi and a webcam) could could get started for just a few pounds. Alternatively, you can buy commercially available kits, 3D printed parts or just hack your own together!
All DIY Robocars rely on the following:
- A scale-model motorised car – radio controlled cars make ideal donor vehicles as most already have servos to control steering, electronic speed control, power, and all the suspension and structures to be resistant to bumps and crashes. Most cars are modified to take the extra stuff you need to bolt on.
- An on-board computer – such as a Raspberry Pi (other boards are also suitable)
- Some sensors – usually a camera, sometimes more including inertia measurement – such as those used to tip your phone screen, an infrared range detector, called a Lidar
- Some software – most people start out using free open-source code which can be downloaded from other projects, then tweaked to suit your vehicle, processor and sensors, and the thing you’re trying to achieve (racing, line following, etc). To get the car to be controlled by the input from a camera, via the on-board computer uses Machine Vision.
- A laptop – so you can easily make changes to the code and upload them to the computer on the car
You don’t need experience. As you’ve landed here, we recommend you read this great introductory blog by Heri Rakotomalala, join the Donkeycar Slack channel, follow @diyrobocars on Twitter, and create an account on Github.
Learn more about the technology behind DIY Robocars
- Inside DonkeyCar (intro to physical AI)
- Supervised Racing (about behavioural cloning vs reinforcement learning)
- Learning to Drive Smoothly with RL
Here are some example builds with instructions:
- Linorobot: LIDAR-guided bot using ROS. Very well documented.
- Minimum Viable Racer – a tiny car with everything you need to get started for less than £100.
- Will Selby’s Elegoo-based rover using variation of FormulaPi code. Very well documented.
- Donkey Car project website
- Build an autonomous car with Raspberry Pi, NAVIO2 and Tensorflow/Keras (Hardware guide) (Software guide)
- Commercially available autonomous racers
- Robocar store, based in Hong Kong – selling the complete Donkey Car kit for about £220
- Active Robots are based nearby in Radstock and sell almost everything you could possibly need via their website.
- Slamtec SDP Mini (Lidar and built-in SLAM processor, around £350)
- Amazon’s DeepRacer (neural network reinforcement learning car, around £400 when it becomes available)
- Zumi (RaspberryPi-based neural network car, around £160 when it launches in summer 2019)
Just because these cars are small and inexpensive doesn’t mean that you can’t run real autonomous car software on them. Like their big brothers, they run ML/CV software, from TensorFlow and OpenCV to ROS and Keras. To keep costs down, they split their computational tasks between Raspberry Pi 3 computers onboard for real-time processing and cloud processing for offline training.
DIY Robocars UK provides:
- a safe friendly environment in which to work on your kits / cars / code
- access to the track
- car parking for your human-driven cars
- a kitchen (and a pizza lunch)
- Wifi access
We do not sell or provide any hardware or software support.