Scholarlink

<p>In this paper, an improved DQN mobile robot path planning algorithm is proposed, which integrates a convolutional neural network with the DQN framework. Firstly, integrating the convolutional model at the outset of the DQN algorithm enables the CNN's output coordinates to accurately reflect the real-time state of the mobile robot. Additionally, a novel greedy exploration mechanism based on an exponential decay function is proposed, aimed at enhancing the exploration probability during the initial stages of the algorithm. Thirdly, to ensure the viability of the shortest path, a reward function based on the endpoint's location is devised. To better tailor the algorithm to the motion characteristics of mobile robots, B-spline curves are introduced for path optimization, producing smoother trajectories. Finally, simulation experiments across various map environments demonstrate that the CNN-DQN and CNN-DQN-B algorithms significantly outperform the traditional DQN algorithm in terms of path optimization, convergence speed, and computational efficiency.</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

Convolutional neural network-based deep Q-network (CNN-DQN) path planning method for mobile robots

  • 期刊论文

  • 发表日期：

    2025年7月1日

查看全文
  • Jiaqiang Zhang,
  • Huiqing Chen,
  • Haoqiang Sun,
  • Hongzhen Xu,
  • Tianxiang Yan

摘要

In this paper, an improved DQN mobile robot path planning algorithm is proposed, which integrates a convolutional neural network with the DQN framework. Firstly, integrating the convolutional model at the outset of the DQN algorithm enables the CNN's output coordinates to accurately reflect the real-time state of the mobile robot. Additionally, a novel greedy exploration mechanism based on an exponential decay function is proposed, aimed at enhancing the exploration probability during the initial stages of the algorithm. Thirdly, to ensure the viability of the shortest path, a reward function based on the endpoint's location is devised. To better tailor the algorithm to the motion characteristics of mobile robots, B-spline curves are introduced for path optimization, producing smoother trajectories. Finally, simulation experiments across various map environments demonstrate that the CNN-DQN and CNN-DQN-B algorithms significantly outperform the traditional DQN algorithm in terms of path optimization, convergence speed, and computational efficiency.