Training Traffic Light Behavior with End-to-End Learning

In this work, we study neural network architectures that will reduce the number of infractions made by autonomous-driving agents. These agents control vehicles providing future waypoints directly from a forward-facing camera. Building on top teacher-student approach Cheating Segmentation, investigate impact Pyramid Pooling Module and Feature Network with aim to learn more representative features. We run our experiment CARLA simulator show pyramid perception modules have positive in reducing traffic light collisions.