Scholarlink

<p>Multi-Agent Path Finding (MAPF) has attracted significant focus in recent times due to its efficiency and scalability, and is widely used in scenarios such as warehouse robots and UAV formations. However, MAPF itself faces many challenges due to its complexity, especially in multi-agent collaboration and conflict resolution. The complexity of MAPF can potentially be addressed using reinforcement learning (RL), but it still faces challenges in scalability. In this paper, we propose a scalable MAPF algorithm based on DHC. Firstly, Coordinate Attention and CondConv are introduced to DHC to construct an enhanced feature extraction module to extract spatial information features observed by agents from both horizontal and vertical directions. Then, we incorporate wavelet transformations into Kolmogorov–Arnold network (KAN) to enhance its ability to handle non-stationary signals. Finally, introducing wavelet transform-based KAN (KAN-WT) to DHC not only improves path planning stability but also minimizes the impact of high-frequency noise and avoid network over-fitting. Experimental results indicate that the performance of the proposed path planning algorithm can be significantly improved under different agent density environments, demonstrating superior scalability and robustness.</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

Scalable multi-agent path planning via wavelet-based KAN and enhanced feature extraction

  • 期刊论文

  • 发表日期：

    2025年6月5日

查看全文
  • Xianchang Liu,
  • Peng Wang,
  • Cui Ni,
  • Hua Wang,
  • Zhu Liu,
  • Haoyuan Shan

摘要

Multi-Agent Path Finding (MAPF) has attracted significant focus in recent times due to its efficiency and scalability, and is widely used in scenarios such as warehouse robots and UAV formations. However, MAPF itself faces many challenges due to its complexity, especially in multi-agent collaboration and conflict resolution. The complexity of MAPF can potentially be addressed using reinforcement learning (RL), but it still faces challenges in scalability. In this paper, we propose a scalable MAPF algorithm based on DHC. Firstly, Coordinate Attention and CondConv are introduced to DHC to construct an enhanced feature extraction module to extract spatial information features observed by agents from both horizontal and vertical directions. Then, we incorporate wavelet transformations into Kolmogorov–Arnold network (KAN) to enhance its ability to handle non-stationary signals. Finally, introducing wavelet transform-based KAN (KAN-WT) to DHC not only improves path planning stability but also minimizes the impact of high-frequency noise and avoid network over-fitting. Experimental results indicate that the performance of the proposed path planning algorithm can be significantly improved under different agent density environments, demonstrating superior scalability and robustness.