An Improved Deep Reinforcement Learning-based Path-planning Algorithm with Dual-experience Pool

摘要

Path planning in unknown environments has been a major challenge for robots. With the recent development of artificial intelligence, deep reinforcement learning has been gradually used to study the path-planning problem of robots in unknown environments due to its autonomous learning characteristics. However, deep reinforcement learning can produce a large amount of redundant data during model training, which reduces the training efficiency. To address this problem, this study proposes a path-planning algorithm based on deep reinforcement learning with a dual-experience pool playback structure. The proposed method realises the segmentation of data using two experience pools and simultaneously adjusts the ratio of experience playback to improve the speed of model training. The proposed algorithm is validated by experiments on a simulation platform using a Turtlebot3 robot. The experimental results show that the proposed algorithm performs better than the commonly used DQN and Double DQN algorithms in terms of convergence speed and accuracy.