AWS DeepRacer is a project by Amazon to encourage more people to develop skills in developing and training neural networks. While their hardware isn’t (yet) available, some of the software techniques used are similar to the DIY Robocars project on which the hardware is based. We asked Kshitija how she found the experience, what she learned, and how she applied this knowledge to building our own DIY autonomous car here at Level Five Supplies.
This was my first experience with DIY autonomous cars – to see one driving autonomously for the first time was incredibly exciting. I met several people at the event working on their own cars and discovered new algorithms that can be used to train neural networks. I also attended an hour-long workshop on reinforcement learning, which is used to train the AWS DeepRacer model.
Building our own robot car
After the event, we ordered the popular Donkey Car hardware kit. It was challenging to build a car from scratch – I had some issues while assembling and configuring Raspberry Pi – but thanks to the Donkey Car Slack community, I was able to make it work. My colleague Flavien and I have been training and testing the Donkey Car on a track we built here at the office.
Our robot car’s Convolutional Neural Network uses a camera to record data and train the model, which enables the car to drive on autopilot mode. We currently use TensorFlow, Python Libraries, NumPy and OpenCV, with Raspberry Pi acting as the computing platform. It costs around £250 to build a DIY autonomous car in this way, so it’s relatively accessible to anyone.
The DIY Robocars UK Meetup organisers expect to run an event in Bath or Bristol in the near future. Join the group for more information and to find out about upcoming events. Alternatively, the AWS DeepRacer community continues to grow – mostly in London.