Keeping an autonomous car from colliding with other vehicles is a challenging challenge to do. Most traditional approaches in this sector depend on model-based solutions, which demand an understanding of vehicle dynamics and an accurate model of vehicle behavior in order to predict the route of the controlled vehicle and surrounding vehicles. It is difficult for these systems to predict and simulate the driving habits of others around them. According to this study, an agent uses deep reinforcement learning to prevent collisions by calculating the distances to neighboring entities and outputting steering angles and accelerations. A variety of roads and cars are included in the Carla Simulator so that the Learning Agent may interact with them and gather data. Such training results in intelligent driving behavior; avoiding crowded areas and traffic situations; reducing speed to minimize rear or frontal accidents; and maneuvering as necessary to avoid side impacts.

Autonomous driving, Deep Reinforcement Learning, Neural Networks, Collision Avoidance, Deep Deterministic Policy Gradients