An Improved Evolutionary Reinforcement Learning Algorithm for UAV Online Target Tracking
Baichen Yu, Feng-Feng Wei, Tai-You Chen, Wei–Neng Chen · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2024
Target tracking for unmanned aerial vehicles (UAVs) is significant in a variety of applications and has high research value. Due to the poor capability of responding to deceptive reward signals and lack of diverse exploration, common reinforcement learning (RL) has limitations in real-time decision-making tasks while evolutionary algorithms (EAs) can compensate for these shortcomings by utilizing fitness metrics and strong exploration ability. In this work, we propose an algorithm named improved evolutionary reinforcement learning (IERL) for online UAV target tracking. Firstly, a realistic UAV online target tracking problem is formulated, considering both the velocity change constraints and sparse reward. Based on this problem formulation, an improved selection operator and an improved interaction setting are proposed to achieve higher individual selection efficiency and higher optimization efficiency, respectively. Simulation results show that the proposed algorithm achieves better performance than the comparison algorithm in the sparse reward tracking tasks.