液压与气动2026，Vol.50Issue(1)：51-60,10.DOI:10.11832/j.issn.1000-4858.2026.01.006
融合强化学习的无模型自适应控制末端执行器恒力控制
Model-free Adaptive Control Based on Reinforcement Learning for End Effectors Constant Force Control
摘要
Abstract
The pneumatic grinding system faces challenges such as nonlinearity,disturbance sensitivity,and modeling issues.The thesis proposes a model-free adaptive control strategy based on reinforcement learning,using the twin delayed deep deterministic policy gradient.This approach is named twin delayed deep deterministic policy gradient-model-free adaptive control.We build a model of the pneumatic loading system in MATLAB/Simulink,design the proposed control strategy,and define state-action and composite reward functions.Simulations show that the proposed control strategy is the optimal controller.Under constant force loading,the proposed control strategy exhibits no significant overshoot and achieves a 0.94 s settling time.Under disturbance loading,the two disturbance peaks are limited to 3.0 N and 3.2 N.For sine force loading,the proposed control strategy tracks the reference with an error of 0.16 N.Under varying angular loading,the proposed control strategy reduces error by 0.4 N.In conclusion,the proposed control strategy exhibits higher control accuracy,faster response,and greater robustness in complex nonlinear pneumatic systems,proving its potential for engineering applications.关键词
气动打磨/无模型自适应控制/强化学习/双延迟深度确定性策略梯度/恒力控制Key words
pneumatic grinding/model-free adaptive control/reinforcement learning/twin delayed deep deterministic policy gradient/constant force control分类
机械制造引用本文复制引用
凌泽懿,张树忠,唐一文,周杰,赵卫..融合强化学习的无模型自适应控制末端执行器恒力控制[J].液压与气动,2026,50(1):51-60,10.基金项目
福建省自然科学基金(2025J01985) （2025J01985）
福建省高校科技创新团队培育计划(闽教科[2020]12号) （闽教科[2020]12号）
