仿人机器人不均匀地面的行走优化算法  

Optimization Algorithm for Walking on Uneven Ground for Humanoid Robots

作　　者：李梦如 

机构地区：[1]苏州科技大学电子与信息工程学院,江苏苏州215009

出　　处：《工业控制计算机》2025年第3期77-79,82,共4页Industrial Control Computer

摘　　要：针对仿人机器人在不均匀地面行走过程中遭遇的内部参数扰动问题,提出了一种融合策略深度强化学习算法。首先,基于关键点匹配技术模仿人类在类似地形上的行走模式为算法提供正向激励,促进策略优化;其次,精细设计马尔科夫决策过程确保行走控制策略的可靠性;最后,基于扩展状态观测器修正对机器人未建模项及环境不确定性的估计,使得机器人的环境感知功能得到提升。实验结果表明,该算法能够控制仿人机器人穿越不均匀的草地地形并保持稳健行走。This paper proposes a fusion strategy deep reinforcement learning algorithm to address the internal parameter disturbance encountered by humanoid robots during uneven ground walking.Firstly,based on key point matching technology,imitating human walking patterns on similar terrains provides positive incentives for the algorithm and promotes strategy optimization.Secondly,finely designing the Markov decision process ensures the reliability of the walking control strategy.Finally,based on the extended state observer correction,the estimation of unmodeled items and environmental uncertainty of the robot is improved,which enhances the robot's environmental perception function.The experimental results show that the algorithm can control humanoid robots to traverse uneven grassland terrain and maintain robust walking.

关 键 词：仿人机器人 行走控制 深度强化学习 扩展状态观测器 运动重定向 

分 类 号：TP2[自动化与计算机技术—检测技术与自动化装置]