TY - GEN
T1 - A Real-time Algorithm for USV Navigation Based on Deep Reinforcement Learning
AU - Zhou, Zhiguo
AU - Zheng, Yipeng
AU - Liu, Kaiyuan
AU - He, Xu
AU - Qu, Chong
N1 - Publisher Copyright:
© 2019 IEEE.
PY - 2019/12
Y1 - 2019/12
N2 - Aiming at the demand of flexibility and real-time performance in unknown aquatorium, a path planning algorithm based on Deep Reinforcement Learning (DRL) is proposed. According to plan-avoid-acclimate request, the proposed algorithm involves optimization of net structure and navigation data enrichment based on A3C, re-regulation of action space of the agent, and is trained with specific tasks in three kinds of maps to improve flexibility. The algorithm is integrated with GPU, which helps achieve high training efficiency and real-time performance by creating a neural network to collect pre-training data. Experimental results show that obstacle ability is confirmed. In comparison with current algorithm, training time reduces by 59.3% and efficiency rises by more than 71.7%. Meanwhile, performance of trained model in unknown environment is validated.
AB - Aiming at the demand of flexibility and real-time performance in unknown aquatorium, a path planning algorithm based on Deep Reinforcement Learning (DRL) is proposed. According to plan-avoid-acclimate request, the proposed algorithm involves optimization of net structure and navigation data enrichment based on A3C, re-regulation of action space of the agent, and is trained with specific tasks in three kinds of maps to improve flexibility. The algorithm is integrated with GPU, which helps achieve high training efficiency and real-time performance by creating a neural network to collect pre-training data. Experimental results show that obstacle ability is confirmed. In comparison with current algorithm, training time reduces by 59.3% and efficiency rises by more than 71.7%. Meanwhile, performance of trained model in unknown environment is validated.
KW - deep reinforcement learning
KW - flexibility
KW - path planning
KW - real-time performance
KW - unmanned surface vehicle
UR - http://www.scopus.com/inward/record.url?scp=85091891243&partnerID=8YFLogxK
U2 - 10.1109/ICSIDP47821.2019.9173280
DO - 10.1109/ICSIDP47821.2019.9173280
M3 - Conference contribution
AN - SCOPUS:85091891243
T3 - ICSIDP 2019 - IEEE International Conference on Signal, Information and Data Processing 2019
BT - ICSIDP 2019 - IEEE International Conference on Signal, Information and Data Processing 2019
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2019 IEEE International Conference on Signal, Information and Data Processing, ICSIDP 2019
Y2 - 11 December 2019 through 13 December 2019
ER -