Reinforcement learning navigation method based on target driving

The invention relates to the technical field of computer vision navigation, in particular to a reinforcement learning navigation method based on target driving. According to the method, the target information in the visual image is extracted more effectively by using the DETR algorithm based on the Transform network, the efficiency is higher, and the generalization is stronger; the designed intelligent agent state feature extraction method enables the reinforcement learning decision network to obtain comprehensive and rich state information to promote network learning; according to the deep reinforcement learning decision network, a heuristic controller is introduced into a baseline PPO algorithm, reinforcement learning training can be helped to converge faster, local optimum even non-convergence is avoided, and the overall efficiency of the algorithm is improved. The method provided by the invention can perform autonomous navigation under the condition of no environmental prior information, and has certain g