Scholarlink

<p>Unmanned aerial vehicles (UAVs), particularly quadcopters, have become increasingly popular due to their versatility, low cost, and ability to operate in remote or hazardous environments. However, their inherently unstable dynamics make autonomous control a significant challenge. In recent years, reinforcement learning has emerged as an efficient alternative to traditional control methods. This paper presents a novel hybrid control strategy that combines proximal policy optimization (PPO) with an improved grey wolf optimizer (IGWO) for precise and stable quadcopter navigation. A deep reinforcement learning-based framework is developed and trained using actor-critic networks. The PPO facilitates faster convergence and more stable learning, while the IGWO enhances global optimization and reduces noise during flight. The proposed method is implemented in a MATLAB-Gazebo simulation environment and is evaluated across various predefined and random trajectories. Performance metrics such as position, velocity, orientation (roll, pitch, yaw), angular rates, motor thrust variations, and cumulative trajectory errors are analyzed. The results demonstrate that the PPO-IGWO approach outperforms traditional optimization and control methods, achieving greater accuracy and robustness. The results are further validated using statistical student t-test where the performance superiority of the proposed model is achieved with at least 95% confidence.</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

Deep reinforcement learning-based autonomous navigation for quadcopter using PPO and improved grey wolf optimizer

  • 期刊论文

  • 发表日期：

    2025年12月19日

查看全文
  • Amit Kumar,
  • Himanshu Yadav,
  • Saksham Nishant Malge,
  • Himanshu Verma,
  • Dakshita Pathak

摘要

Unmanned aerial vehicles (UAVs), particularly quadcopters, have become increasingly popular due to their versatility, low cost, and ability to operate in remote or hazardous environments. However, their inherently unstable dynamics make autonomous control a significant challenge. In recent years, reinforcement learning has emerged as an efficient alternative to traditional control methods. This paper presents a novel hybrid control strategy that combines proximal policy optimization (PPO) with an improved grey wolf optimizer (IGWO) for precise and stable quadcopter navigation. A deep reinforcement learning-based framework is developed and trained using actor-critic networks. The PPO facilitates faster convergence and more stable learning, while the IGWO enhances global optimization and reduces noise during flight. The proposed method is implemented in a MATLAB-Gazebo simulation environment and is evaluated across various predefined and random trajectories. Performance metrics such as position, velocity, orientation (roll, pitch, yaw), angular rates, motor thrust variations, and cumulative trajectory errors are analyzed. The results demonstrate that the PPO-IGWO approach outperforms traditional optimization and control methods, achieving greater accuracy and robustness. The results are further validated using statistical student t-test where the performance superiority of the proposed model is achieved with at least 95% confidence.