The internship consist of a group of motivated highschoolers interested in deep learning. We all had robotics experience from our FRC teams and had worked with computer vision.
The racecar was developed by JetsonHacks, modelled after the MIT RACECAR (Competing Ackermann-steering Robot). The car has on board: a Jetson TX2 ,2D Lidar - Hokuyo UST-10LX and a Microsoft LifeCam.
We used Keras built on top of Tensorflow to design the neural networks for steering. There are two individual networks, one for turning left and one that turns right. A pathfinding algorithm toggles each of these networks on and off. For finding people, we used Caffe and a fine-tuned GoogleNet with custom pictures of our legs (because the car is short).