IRJET - Obstacle Detection using a Stereo Vision of a Car

This document describes a study that uses deep reinforcement learning to build a self-driving car agent. The agent takes raw sensory inputs from a simulation environment and learns to navigate and control the car through a deep Q-network trained with Q-learning. The authors implement the self-driving car using PyTorch and evaluate its performance against other standard agents. The results show that their agent is able to successfully control the car to navigate within the simulation environment.