Performance-Designated Reinforcement Learning Enclosing Control for UAVs With Collision-Free Capability

作者：一平 山下, Xingling Shao, Zewei Mei, Wendong Zhang · 发表于：IEEE Transactions on Intelligent Transportation Systems · 年份：2024 · DOI：10.1109/tits.2024.3384431 · 被引用次数：25 · 研究领域：Robotic Path Planning Algorithms、Autonomous Vehicle Technology and Safety、Traffic control and management

Target enclosing is a great potential avenue for unmanned aerial vehicles (UAVs) to execute search and rescue, escorting, and geography mapping missions. However, existing methods struggle to fulfill the requirements of prescribed enclosing performance and collision avoidance in a constrained environment. To address aforementioned issues, this paper aims to propose a performance-designated reinforcement learning-based enclosing control (PDRLEC) scheme for UAVs to achieve target circle approximation with specified manners while ensuring collision avoidance. Particularly, an adaptive variable performance function is designed to effectively address the singularity problem existing in the prescribed performance control (PPC). Additionally, by converting radial deviation and distance between UAVs and obstacles into two skilled barrier functions, the reinforcement learning (RL) module is endowed with the ability to handle performance constraints and achieve obstacle bypassing. The proposed PDRLEC, which consists of a robust enclosing item and a RL-related module, can achieve optimal enclosing with enhanced sample efficiency and elevated reward values while striking a balance between collision avoidance and target enclosure. Finally, abundant simulations are provided to corroborate the feasibility and superiority of developed PDRLEC.