We have been hard at work to create (to our knowledge) the world’s first fully online learning self-driving mini-car!
Using a stock RC car model, we equipped it with a Raspberry Pi 3 along with an Arduino to control the servos/speed controller. We used the latest iteration of our online learning software, EOgmaNeo, to provide the brains of the car.
All inference and learning is done right on the Raspberry Pi’s CPU, no pre-training, no connection to a more powerful machine. The car uses a PiCamera to provide visual inputs and a steam controller to provide steering targets when in training mode.
The model is able to run in real-time with ~10 million synapses at 60 frames per second on the Pi.
Stay tuned for more information and a source code release!
Here is a video of the car in action!