Here for a quick update on the state of the car, platform and Tensorflow/Keras models I’m currently building and testing in various environments. I’ve been working on lately on an indoor model to drive a car around the dining table in our home. It is a simple oval track, but it is visually cluttered and quite tight. Here’s a video showing a couple of successful laps, and some where the car gets stuck as well:
The driving platform that the car uses is Burro – A project that started as a fork of Donkey, and lately it’s kind of following it’s own path. The Keras model is evaluated onboard, and reaches around 10-12fps. This is a Raspberry PI 2 with no overclocking. I’d like to try overclocking as well, but I have a couple more thoughts I want to try to improve fps before I go on with it.
The model is trained using Keras/Tensorflow on a dataset of around 18000 images. There are no sensors onboard other than vision using a wide-angle camera. No ultrasound/LIDAR/TOF here at all1. I’m planning to write up an extensive post outlining the process of data collection (including strategies for driving), pre-processing and configuring training parameters. I’ve also developed a pipeline of Python generators for preprocessing and transforming images, including rotations, mirroring, color adjustments and other operators, which I’d like to share in the near future.
So stay tuned and don’t forget to share!