Research on path planning algorithm based on deep reinforcement learning

Replacing repetitive and uncreative operations through Robotic Process Automation (RPA) to reduce labor cost and task completion time has been the mega trend. In this paper, we combine a deep learning model and an augmented learning model, using the augmented learning model to interact with the environment and the deep learning model to assist in recording the explored paths, to realize the collection of complete execution trajectories from the production environment. Aiming at the problems of path planning, this paper proposes an RPA path planning method based on the improved Double Deep Q-learning algorithms (DDQN) algorithm, and proposes the corresponding solutions and realizes them based on the problems, and finally proves the effectiveness and feasibility of the algorithm. In this paper, a new sampling strategy is proposed, which aims to improve the exploration efficiency of the DDQN algorithm. The method is based on the improvement of the ranking priority sampling strategy, which takes the Jaccard coefficient of the location information between samples as the sample priority, and improves it along the importance sample collection strategy, so as to eliminate the problems of the importance sampling method.