Scholarlink

<p>This paper presents a comprehensive comparative study of advanced Deep Deterministic Policy Gradient (DDPG) variants for trajectory tracking control of a 5-degree-of-freedom (DOF) Mitsubishi RV-2AJ robotic arm. The investigation centers on three reinforcement learning algorithms: Classical DDPG, Lyapunov-Constrained DDPG (LC-DDPG), and Twin Delayed DDPG with Adaptive Exploration (TD3-ADX). These model-free, actor-critic-based approaches are evaluated against traditional control strategies such as Proportional-Integral-Derivative (PID) and Adaptive Neuro-Fuzzy Inference System (ANFIS), to benchmark performance in terms of accuracy, stability, and learning efficiency. The robotic system is modeled in MATLAB/Simulink with Simscape Multibody, and the control agents are trained using a reward function inspired by artificial potential fields to encourage smooth, energy-efficient, and precise trajectory tracking. The classical DDPG serves as the baseline, LC-DDPG enhances stability and robustness through Lyapunov-based safety constraints, and TD3-ADX mitigates overestimation bias and accelerates learning through adaptive exploration mechanisms. Simulation results demonstrate that the advanced DDPG variants outperform classical PID and ANFIS controllers, particularly in dynamic and uncertain environments. Among them, LC-DDPG excels in maintaining system stability under disturbances, whereas TD3-ADX achieves superior tracking precision and faster convergence. This study underscores the potential of next-generation reinforcement learning algorithms in achieving high-performance, autonomous control for robotic manipulators, and paving the way for more intelligent and resilient robotic systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Reinforcement learning-based intelligent trajectory tracking for a 5-DOF Mitsubishi robotic arm: comparative evaluation of DDPG, LC-DDPG, and TD3-ADX

  • 期刊论文

  • 发表日期：

    2025年7月10日

查看全文
  • Zied Ben Hazem,
  • Firas Saidi,
  • Nivine Guler,
  • Ali Husain Altaif

摘要

This paper presents a comprehensive comparative study of advanced Deep Deterministic Policy Gradient (DDPG) variants for trajectory tracking control of a 5-degree-of-freedom (DOF) Mitsubishi RV-2AJ robotic arm. The investigation centers on three reinforcement learning algorithms: Classical DDPG, Lyapunov-Constrained DDPG (LC-DDPG), and Twin Delayed DDPG with Adaptive Exploration (TD3-ADX). These model-free, actor-critic-based approaches are evaluated against traditional control strategies such as Proportional-Integral-Derivative (PID) and Adaptive Neuro-Fuzzy Inference System (ANFIS), to benchmark performance in terms of accuracy, stability, and learning efficiency. The robotic system is modeled in MATLAB/Simulink with Simscape Multibody, and the control agents are trained using a reward function inspired by artificial potential fields to encourage smooth, energy-efficient, and precise trajectory tracking. The classical DDPG serves as the baseline, LC-DDPG enhances stability and robustness through Lyapunov-based safety constraints, and TD3-ADX mitigates overestimation bias and accelerates learning through adaptive exploration mechanisms. Simulation results demonstrate that the advanced DDPG variants outperform classical PID and ANFIS controllers, particularly in dynamic and uncertain environments. Among them, LC-DDPG excels in maintaining system stability under disturbances, whereas TD3-ADX achieves superior tracking precision and faster convergence. This study underscores the potential of next-generation reinforcement learning algorithms in achieving high-performance, autonomous control for robotic manipulators, and paving the way for more intelligent and resilient robotic systems.