The goal of this project is to further train and deploy a neural network to a remote control car that is 1/16th scale that will enable it to drive autonomously around a racetrack with at least one other car on the track. The Donkey Car will function as close to a human driver as possible because the neural network will be trained with data generated by a human driver which was possible with the assistance of the open source Donkey Car project. Additionally, the car will be trained to develop a passing behavior allowing the model to increase the throttle and shift lanes, to effectively pass the slower car.

The video below showcases the successful passing behavior of the autonomous RC car. The model for the red car identifies the gray car in front of the driver and as it gets closer it decides to shift to the left lane. While this is happening, the speed of the gray car is calculated. Then the steering of the red car turns left and the speed increases to get into the left lane. Depending on the red car’s prediction for speed and position in the track of the gray car, the model identifies when to shift back into the left lane. It is important to note that the model of the red car is able to do this after the gray car disappears from view as the only sensor is one camera on the front of the car. This has led to a dependence on a robust model trained for autonomous control.