Study on the Path Planning Algorithm Based on Dueling Deep Q Network

In traditional path planning methods based on Deep Q-Network (DQN) algorithm for mobile robots, the target Q value was usually obtained by a greedy algorithm, and the estimation of Q value was usually high, which would result in slow training speed of the algorithm. To solve this problem, a Dueling...

Full description

Saved in:
Bibliographic Details
Published in:Journal of physics. Conference series 2021-05, Vol.1920 (1), p.12084
Main Authors: Cao, Jingwei, Zhao, Shancheng, Guo, Caiqin, Hu, Youtao, Gao, Qianxu, Shen, Fengjun, Ye, Yujie, Zhang, Yong
Format: Article
Language:English
Subjects:
Algorithms
Convergence
Decision making
Greedy algorithms
Neural networks
Optimization
Path planning
Physics
Training
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In traditional path planning methods based on Deep Q-Network (DQN) algorithm for mobile robots, the target Q value was usually obtained by a greedy algorithm, and the estimation of Q value was usually high, which would result in slow training speed of the algorithm. To solve this problem, a Dueling DQN algorithm based on Dueling network was proposed in this paper, which was mainly realized by the structure optimization of neural network. The initial algorithm structure was divided into two states and a higher return is obtained through competitive comparison. Then, the long training time and slow speed convergence problem could be effectively solved than the traditional DQN. The simulation results show that the DQN algorithm based on Dueling DQN has high efficiency in Q value obtaining. Compared with the traditional DQN method, it has the better anti-interference ability and decision-making ability, less training time, and stable convergence.
ISSN:1742-6588
1742-6596
DOI:10.1088/1742-6596/1920/1/012084