Scholarlink

<p>Mobile robot control in pursuit–evasion scenarios poses significant challenges due to rapidly changing environments, unpredictable adversaries, and the need for real-time decision-making. Existing approaches often struggle to adapt dynamically while ensuring optimal performance regarding evasion success, resource efficiency, and safety. To solve this problem, we propose a novel Adaptive Model Predictive Control (A-MPC) framework integrated with Deep Reinforcement Learning (DRL) and differential game theory. A-MPC distinguishes itself by dynamically adjusting control actions in response to real-time environmental changes, allowing mobile robots to predict adversarial behavior and optimize their strategies continuously. Double integrator dynamics ensures smooth control over position and velocity, while an adaptive reward structure balances critical objectives like evasion, energy efficiency, and collision avoidance. Deep Q-Network (DQN) techniques are also used in a feedback loop to refine decision-making, enhancing learning efficiency and robustness. The novelty of this approach lies in the seamless integration of A-MPC with DRL, enabling precise, adaptive maneuvers in complex multi-agent environments. Experimental evaluations demonstrate significant improvements in pursuit–evasion tasks, including reduced capture time, enhanced resource utilization, and superior agent coordination. This framework offers a comprehensive solution for real-time, intelligent evasion strategies, making it ideal for applications such as autonomous drones and robotic systems requiring high adaptability in dynamic environments.</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

Mobile Robot Control Using Pursuit–Evasion Differential Game Strategy for Double Integrator Dynamic Control with Deep Reinforcement Learning

  • 期刊论文

  • 发表日期：

    2025年5月8日

查看全文
  • Li Zhenxiang,
  • Mrim M. Alnfiai,
  • Low Tang Jung,
  • Nouf Nawar Alotaibi,
  • Mohd Yazid Bajuri,
  • Salma Mohsen M. Alnefaie,
  • Mohamed E. Ghoneim

摘要

Mobile robot control in pursuit–evasion scenarios poses significant challenges due to rapidly changing environments, unpredictable adversaries, and the need for real-time decision-making. Existing approaches often struggle to adapt dynamically while ensuring optimal performance regarding evasion success, resource efficiency, and safety. To solve this problem, we propose a novel Adaptive Model Predictive Control (A-MPC) framework integrated with Deep Reinforcement Learning (DRL) and differential game theory. A-MPC distinguishes itself by dynamically adjusting control actions in response to real-time environmental changes, allowing mobile robots to predict adversarial behavior and optimize their strategies continuously. Double integrator dynamics ensures smooth control over position and velocity, while an adaptive reward structure balances critical objectives like evasion, energy efficiency, and collision avoidance. Deep Q-Network (DQN) techniques are also used in a feedback loop to refine decision-making, enhancing learning efficiency and robustness. The novelty of this approach lies in the seamless integration of A-MPC with DRL, enabling precise, adaptive maneuvers in complex multi-agent environments. Experimental evaluations demonstrate significant improvements in pursuit–evasion tasks, including reduced capture time, enhanced resource utilization, and superior agent coordination. This framework offers a comprehensive solution for real-time, intelligent evasion strategies, making it ideal for applications such as autonomous drones and robotic systems requiring high adaptability in dynamic environments.