Dueling DQN-Rollout for Collision Avoidance Path Planning with Vehicle Speed Location

Abstract

The rapid progress of artificial intelligence has led to significant advancements in the field of autonomous driving, yet effective collision avoidance path planning remains a challenging task. In response, deep reinforcement learning offers an efficient and modern alternative to traditional navigation strategies. This paper proposes a novel approach that incorporates vehicle speed location into the deep reinforcement learning process, utilizing the Dueling DQN-Rollout framework to consider both the distance of the road and obstacles ahead. The agent interacts with the environment to learn a policy, with a reward function that accounts for deviations from the intended path and collisions with obstacles. The training process focuses on imparting human-like driving skills to the autonomous vehicle. By employing the rollout algorithm, the rough Q-value is optimized to reduce training costs. Experimental results demonstrate that this approach can successfully plan a collision-free path for autonomous driving from origin to destination on a simulation platform.