Search

Aiming at the confrontation game problems between pursuit-evasion unmanned surface vehicles under complex multi-obstacle environment, a pursuit-evasion game strategy is proposed. Firstly, the multi-obstacle environment is set up, and the gaming situation can be judged by the perception between pursuit-evasion USVs. For the pursuers, the model training is performed based on multi-agent deep reinforcement learning, so that they can quickly plan a reasonable obstacle avoidance and pursuit route, and form an effective encirclement posture before the evader approaches the target point. Meanwhile, the credit assignment problem among the members of the pursuing group is considered. For the evader, deep reinforcement learning is combined with imitation learning to train the escape model, so that it can reach the preset point in as short a time as possible and avoid the obstacles on the way. Finally, an adversarial-evolutionary game training method under multiple random scenarios is designed and combined with curriculum learning to iteratively update the pursuit and escape models. Through the detailed comparative analysis of the model training process and simulation experiments, it is proved that the proposed two types of models have higher convergence efficiency and stability, and they can have higher intelligence to pursue, escape and avoid obstacles respectively. • A pursuit-evasion strategy is proposed for USVs under complex obstacles environment. • Based on DRL, flexible rewards guide pursuers form stable encirclement. • Combined DRL and IL, escape strategies are trained to make a evader intelligent. • An adversarial-evolutionary training method is designed to iteratively update models. [ABSTRACT FROM AUTHOR]

This paper studies a guidance and control framework of multiple autonomous surface underwater vehicles (multi-ASUV) based on deep reinforcement learning (DRL) for target tracking. The framework enables the vehicles to complete the standoff tracking and sampling tasks in a predetermined circular trajectory centered on the target and maintain predetermined relative positions during the process to obtain high-precision spatio-temporal synchronization data. We design an end-to-end architecture that maps the sensor inputs to control commands and develop autonomous capable of achieving the hybrid objective of cooperative guidance, standoff tracking, and dynamic obstacle avoidance without having prior knowledge about the goal or the environment. The results demonstrate the feasibility of the end-to-end DRL method with higher accuracy than the traditional "guidance-control" two-step method. Meanwhile, the obstacle avoidance and standoff tracking experiment for swarm and the sampling experiment in the mesoscale eddy area are stimulated further to verify the proposed framework's effectiveness and robust ability. • We propose a Deep-Reinforcement-Learning-based Multi-vehicle Cooperative Standoff Tracking Framework. • Through end-to-end control, each vehicle eventually converges to the tracking circle and distributes evenly. • This paper solve the difficult-to-eliminate uncertainty and delay caused by the combination of "guidance-control" two steps method. [ABSTRACT FROM AUTHOR]