Research on decision control of unmanned driving based on SAC

An improved method called SACwTS (Soft Actor-Critic with Twice Sampling) coupled with twice sampling is presented in order to improve defect with the Soft Actor-Critic (SAC) algorithm, such as the waste of important sample resources, long training times, and unstable training process. The algorithm is based on the idea of replacing uniform sampling with non-uniform sampling through prioritized experience replay. As a result, it first chooses the sequences with large cumulative returns in the sequence distribution calculated according to the cumulative returns with high probability, and then it chooses the sequences with higher TD-error (Temporal-difference error) priority according to the prioritized experience replay to finish the twice-sampling of the sample sequence data. The comparison of this algorithm’s quality with that of the fundamental SAC algorithm when applied to the unmanned TORCS simulation platform makes it abundantly evident that this algorithm is superior.