基于深度强化学习的液压四足机器人单腿跳跃优化控制  

Optimization Control of Hydraulic Quadruped Robot Single Leg Jumping Based on Deep Reinforcement Learning

作　　者：姬鹏远 纵怀志 艾吉昆 张军辉[1] 徐兵[1] JI Pengyuan;ZONG Huaizhi;AI Jikun;ZHANG Junhui;XU Bing(State Key Laboratory of Fluid Power&Mechatronic Systems,Zhejiang University,Hangzhou,Zhejiang 310027)

机构地区：[1]浙江大学流体动力与机电系统国家重点实验室,浙江杭州310027

出　　处：《液压与气动》2025年第1期68-75,共8页Chinese Hydraulics & Pneumatics

摘　　要：液压四足机器人具有功率密度高、负载能力大等优势,但其液压系统控制参数与机身运动参数间耦合关系复杂且维度较高,导致最优控制参数的求解十分困难。为此,采用具有自适应性、抗扰动性的深度强化学习算法,实现机器人单腿在不同工况下的高效、平稳运动。首先,在MATLAB/Simulink中搭建Spurlos液压四足机器人单腿模型;然后,设计基于五次多项式轨迹规划的强化学习控制器;最终实现针对不同目标任务的机器人单腿优化控制。仿真表明,所提强化学习控制策略能够实现机器人跳跃运动的自适应优化控制,训练后的机器人单腿在复杂环境中展现出较强的自适应性与运动稳定性。Hydraulic quadruped robots possess advantages such as high power density and large load capacity.However,due to the complex and high-dimensional coupling relationship between the control parameters of their hydraulic systems and the body motion parameters,it is extremely difficult to solve for the optimal control parameters.To address this issue,this paper employs a deep reinforcement learning algorithm with adaptability and disturbance resistance to achieve efficient and smooth motion of a robot's single leg under different operating conditions.Initially,a Spurlos hydraulic quadruped robot single leg model is constructed in MATLAB/Simulink.Subsequently,a reinforcement learning controller based on fifth-order polynomial trajectory planning is designed.Ultimately,this enables optimized control of the robot's single leg for various target tasks.Simulations show that the proposed reinforcement learning control strategy achieves adaptive optimization of robot jumping motion.The trained robot leg exhibits strong adaptability and motion stability in complex environments.

关 键 词：液压四足机器人 深度强化学习 跳跃控制 轨迹规划 

分 类 号：TH137[机械工程—机械制造及自动化]