TY - GEN
T1 - Application of Actor-Critic Deep Reinforcement Learning Method for Obstacle Avoidance of WMR
AU - Gao, Xiaoshan
AU - Yan, Liang
AU - Wang, Gang
AU - He, Zhuang
AU - Gerada, Chris
AU - Chang, Suokui
N1 - Publisher Copyright:
© 2022, The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
PY - 2022
Y1 - 2022
N2 - A state-of-the-art framework, i.e., deep deterministic policy gradient (DDPG), has obtained a certain effect in the robotic control field. When the wheeled mobile robot (WMR) executes operation in unstructured environment, it is critical to endow the WMR with the capacity to avoid the static and dynamic obstacles. Thus, a obstacle avoidance algorithm based on DDPG is proposed to realize the autonomous navigation in the unknown environment. The WMR in this study installs the requisite sensors to provide the fully observable environment information at any moment. The continuous state space description for WMR and obstacles is designed, together with the reward mechanism and action space. The learning agent. i.e., the studied mobile robot, utilizes the DDPG model, through the continuous interaction with the surrounding environment and the application of historical experience data, the WMR can learn the optimal action behavior. Simulation along with test works strongly verify the collision-free ability in static and dynamic scenarios with multiple observable obstacles.
AB - A state-of-the-art framework, i.e., deep deterministic policy gradient (DDPG), has obtained a certain effect in the robotic control field. When the wheeled mobile robot (WMR) executes operation in unstructured environment, it is critical to endow the WMR with the capacity to avoid the static and dynamic obstacles. Thus, a obstacle avoidance algorithm based on DDPG is proposed to realize the autonomous navigation in the unknown environment. The WMR in this study installs the requisite sensors to provide the fully observable environment information at any moment. The continuous state space description for WMR and obstacles is designed, together with the reward mechanism and action space. The learning agent. i.e., the studied mobile robot, utilizes the DDPG model, through the continuous interaction with the surrounding environment and the application of historical experience data, the WMR can learn the optimal action behavior. Simulation along with test works strongly verify the collision-free ability in static and dynamic scenarios with multiple observable obstacles.
KW - Deep reinforcement learning
KW - Obstacle avoidance
KW - Wheeled mobile robot
UR - http://www.scopus.com/inward/record.url?scp=85120634140&partnerID=8YFLogxK
U2 - 10.1007/978-981-15-8155-7_453
DO - 10.1007/978-981-15-8155-7_453
M3 - Conference contribution
AN - SCOPUS:85120634140
SN - 9789811581540
T3 - Lecture Notes in Electrical Engineering
SP - 5485
EP - 5494
BT - Advances in Guidance, Navigation and Control - Proceedings of 2020 International Conference on Guidance, Navigation and Control, ICGNC 2020
A2 - Yan, Liang
A2 - Duan, Haibin
A2 - Yu, Xiang
PB - Springer Science and Business Media Deutschland GmbH
T2 - International Conference on Guidance, Navigation and Control, ICGNC 2020
Y2 - 23 October 2020 through 25 October 2020
ER -