Improved RRT Path Planning Algorithm Based on Deep Q-network
LI Zhaoying1， OU Yiming2， SHI Ruoling1
1. School of Astronautics， Beihang University， Beijing 100191， China； 2. Department of Mechanical Engineering and Automation， Harbin Institute of Technology， Shenzhen 518055， Guangdong， China
Abstract Aiming at the problems of large randomness and low search efficiency of rapid exploring random tree (RRT) path planning algorithm, combined with the characteristics that reinforcement learning can select strategies according to prior knowledge, an improved RRT optimization algorithm based on deep Q-network (DQN) is proposed. Firstly, the obstacle avoidance strategy with variable step in complex domain is designed, and the Markov decision process (MDP) model of random tree growth in RRT algorithm is established. Then, the obstacle avoidance strategy and MDP model are connected to the interface of RRT-Connect algorithm, and the specific process of training and path planning is designed. Finally, the simulation experiment is carried out on the MATLAB software platform. The simulation results show that the improved RRT-Connect algorithm based on deep Q-network (DQN-RRT-C) has a significant improvement in rapidity and search efficiency.
2021, Vol. 4 
