DQN-based Reinforcement Learning for Vehicle Control of Autonomous Vehicles Interacting With Pedestrians

Autonomous Vehicles (AVs) have become a popular research topic in recent years due to their ability to improve road safety by reducing traffic accidents and human injuries. Vehicle control is the most significant part of autonomous driving, which adjusts the steering angle and velocity of AVs during driving. Recently, vehicle control has seen consequential progress using effective Artificial Intelligence (AI), especially Deep Learning (DL) techniques. Recent works have been limited to using Reinforcement Learning (RL) techniques to control AV to follow only its path without taking into consideration other road users, especially the pedestrians. In this paper, we propose a Novel Reinforcement Learning based model using Deep-Q Networks to control the AV in a complex scenario involving vehicles and pedestrians. AV learns the policy of several actions in order to reach its destination without accidents with other road participants. Our approach in tested and validated using the CARLA simulator. Our results show that the proposed approach achieves better performances in terms of average reward, success rate, and collision rate over time.