Pedestrian collision avoidance using deep reinforcement learning

One of the main challenges in transportation is the high fatality rate caused by vehicle-pedestrian collisions. This issue is exacerbated by a variety of abnormal and unpredictable situations. This study proposes a novel smart algorithm for pedestrian collision avoidance based on deep reinforcement learning. A deep Q-network (DQN) is designed to learn an optimal driving policy for pedestrian collision avoidance across diverse environments and conditions. The algorithm interacts with both vehicle and pedestrian agents and employs a new reward function to train the model. We used Car Learning to Act (CARLA), an open-source autonomous driving simulator, to train and validate the model under various conditions. Applying the proposed algorithm in a simulated environment reduced vehicle-pedestrian collisions by approximately 64%, depending on the specific conditions. Our findings offer an early-warning solution to reduce the risk of vehicle-pedestrian collisions in real-world scenarios.