Application of DDPG algorithm with fused experience pool for autonomous driving

The development of smart transportation and the popularization of artificial intelligence had to the widespread use of autonomous driving technology. Among the various algorithms used in this field, the deep deterministic policy gradient algorithm (DDPG) has gained popularity among researchers., The uniform sampling approach used in the DDPG algorithm can lead to several issues, including a high proportion of inefficient strategies, easy discarding of high reward samples, low training efficiency, and slow convergence speed. To overcome these problems, this study proposes a deep deterministic policy gradient (E-DDPG) algorithm that incorporates a fusion experience pool. By establishing a successful sample experience pool, the algorithm improves the utilization rate of sampling, enhances the exploration strategy, and increases the efficiency of neural network training. The performance of the E-DDPG algorithm was evaluated on the TORCS racing simulation platform, demonstrating its ability to enhance the learning efficiency of driving strategies in autonomous driving tasks compared to existing algorithms