Open Access Open Access  Restricted Access Subscription Access

COLREGs-compliant dynamic collision avoidance algorithm based on deep deterministic policy gradient


Affiliations
1 School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai – 200 093, China
2 Mechanical Engineering College, Beihua University, Jilin – 132 021, China
3 Department of EE Data Architecture, Human Horizons Technology Co., Ltd., Shanghai – 200 082, China

In order to reduce collision avoidance accidents and improve the safety of ship navigation, a dynamic collision avoidance algorithm based on deep reinforcement learning is proposed in this paper. In order to avoid the fuzziness and uncertainty in the encounter process, the degree of risk is formulated to quantify the collision risk. International regulations for preventing collisions at sea (COLREGs) are quantified reasonably. Considering the factors of collision, position, speed, course and compliance with the COLREGs, the reward function of the algorithm is designed to ensure that the collision avoidance decision is safe and effective and meet the requirements of the COLREGs. Based on DDPG algorithm, the sample data processing mechanism is improved, the utilization rate of experience is improved, and the problems of long learning time and unstable training are solved. The navigation and collision avoidance for multiple ships are simulated respectively. The results show that this method can effectively avoid obstacle ships under the requirements of COLREGs, and it has good real-time performance and safety.
User
Notifications
Font Size

Abstract Views: 92




  • COLREGs-compliant dynamic collision avoidance algorithm based on deep deterministic policy gradient

Abstract Views: 92  | 

Authors

X L Xu
School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai – 200 093, China
X L Zhou
Mechanical Engineering College, Beihua University, Jilin – 132 021, China
P Cai
Department of EE Data Architecture, Human Horizons Technology Co., Ltd., Shanghai – 200 082, China
Z Z Chu
School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai – 200 093, China

Abstract


In order to reduce collision avoidance accidents and improve the safety of ship navigation, a dynamic collision avoidance algorithm based on deep reinforcement learning is proposed in this paper. In order to avoid the fuzziness and uncertainty in the encounter process, the degree of risk is formulated to quantify the collision risk. International regulations for preventing collisions at sea (COLREGs) are quantified reasonably. Considering the factors of collision, position, speed, course and compliance with the COLREGs, the reward function of the algorithm is designed to ensure that the collision avoidance decision is safe and effective and meet the requirements of the COLREGs. Based on DDPG algorithm, the sample data processing mechanism is improved, the utilization rate of experience is improved, and the problems of long learning time and unstable training are solved. The navigation and collision avoidance for multiple ships are simulated respectively. The results show that this method can effectively avoid obstacle ships under the requirements of COLREGs, and it has good real-time performance and safety.