A Donkey car is an open source DIY self driving platform powered by a Raspberry Pi running a Deep Learning network. In this talk we'll go into how we modified the platform to retrain the Deep Learning network on a AWS GPU instance.
The limited compute power of a Raspberry Pi severely hinders its capability to run and train a deep learning network. However, there are options to improve the available compute. We choose an AWS GPU instance to retrain the model, but had to recompile tensorflow for arm in the process to allow it to be run on the raspberry.