空间控制技术与应用2025，Vol.51Issue(5)：1-15,15.DOI:10.3969/j.issn.1674-1579.2025.05.001
基于深度强化学习的月面四足机器人控制方法综述
A Review of Deep Reinforcement Learning-Based Control Methods for Lunar Quadruped Robots
摘要
Abstract
With the evolution of lunar exploration missions transitioning from mainly exploration to integrated resource exploration and utilization,lunar surface mobility equipment urgently necessitates enhanced terrain adaptability and autonomous decision-making capabilities.Quadruped robots,leveraging discrete foot-end support mechanisms,multi-degree-of-freedom joint architectures,and modular design paradigms,demonstrate superior obstacle avoidance capabilities.This enables access to complex terrains such as crater bottoms and lava tubes,which prove challenging for wheeled robotic systems.Nevertheless,the extreme lunar environment imposes rigorous demands on robotic motion control,imperatively requiring real-time autonomous decision-making capacities.Classic control methodologies,reliant on precise dynamic models,struggle to accommodate abrupt variations in lunar regolith mechanics and terrain uncertainty.In this context,deep reinforcement learning(DRL)offers a viable solution through its end-to-end perception-decision-execution closed-loop framework,effectively addressing lunar soil mechanical uncertainties and providing sustainable reliability for extended lunar missions.DRL-based control methodologies have emerged as a focal research area in quadruped robot control in recent years.This paper aims to systematically review the advantages and challenges of DRL-enabled quadruped robots in lunar exploration.The current research landscape of classical control methods for quadruped robots is examined.The developmental trajectory,strengths,and limitations of mainstream DRL algorithms are delineated.Following this,the paper discusses recent advancements in DRL-based control algorithms for lunar quadruped robots.Finally,the predominant challenges and prospective research directions are synthesized for implementing DRL in lunar quadruped robotic systems.关键词
深度强化学习/四足机器人/月球探测/无模型控制Key words
deep reinforcement learning/quadruped robots/lunar exploration/model-free control分类
航空航天引用本文复制引用
唐力强,韦宇轩,胡海东,胡勇,杨孟飞..基于深度强化学习的月面四足机器人控制方法综述[J].空间控制技术与应用,2025,51(5):1-15,15.基金项目
国家自然科学基金资助项目(62193273020)和国家重点研发计划(2024YFB3909905) National Natural Science Foundation of China(62193273020)and National Key Research and Development Program(2024YFB3909905) （62193273020）
