Scholarlink

<p>The navigation of mobile autonomous robots in changing environments requires path planning in a safe, smooth, and adaptable manner. Traditional optimization, sampling, and learning-based approaches mitigate individually processed attributes but fail to assure scalability, resilience, and intent-aware coordination jointly. This research proposes a Collaborative Graph-based Navigation and Intent-aware Robots (COGNIRobots), a comprehensive approach that integrates multisensory cross-attention fusion, episodic memory using memformer, a hybrid Revobuilder optimizer, Intent-Aware Graph Attention Network, and Collaborative Gated Bidirectional Long-Short Term Memory-based trajectory generation to address the limitations of the traditional techniques. This suggested framework is assessed against a number of state-of-the-art approaches under four test environments with a lot of obstacles. According to the experimental results, COGNIRobot significantly outperforms traditional baseline methods in terms of proficiency and reliability. It achieves an accuracy of 97.3% with a unique intersection over union (MIoU) of 0.81, provides an average of 12–15% shorter paths, reduces collisions to a rate of 3.5%, and reduces overall cost to an average of 0.95. Additionally, compared to conventional baselines, the COGNIRobots framework achieved a significantly lower average path variance (0.21) and average path delay (0.95&#xa0;s), demonstrating its resilience in various scenarios. The outcomes demonstrated the potential of integrating memory-based adaptation, multimodality integration, and intent-aware coordination to facilitate multi-robot navigation in realistic 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

A multi-agent contextual memory-driven optimization framework with deep learning and intent-aware coordination for mobile robot path planning

  • 期刊论文

  • 发表日期：

    2026年5月19日

查看全文
  • Dinesh Selvaraj,
  • V. S. Winstor Jebakumar

摘要

The navigation of mobile autonomous robots in changing environments requires path planning in a safe, smooth, and adaptable manner. Traditional optimization, sampling, and learning-based approaches mitigate individually processed attributes but fail to assure scalability, resilience, and intent-aware coordination jointly. This research proposes a Collaborative Graph-based Navigation and Intent-aware Robots (COGNIRobots), a comprehensive approach that integrates multisensory cross-attention fusion, episodic memory using memformer, a hybrid Revobuilder optimizer, Intent-Aware Graph Attention Network, and Collaborative Gated Bidirectional Long-Short Term Memory-based trajectory generation to address the limitations of the traditional techniques. This suggested framework is assessed against a number of state-of-the-art approaches under four test environments with a lot of obstacles. According to the experimental results, COGNIRobot significantly outperforms traditional baseline methods in terms of proficiency and reliability. It achieves an accuracy of 97.3% with a unique intersection over union (MIoU) of 0.81, provides an average of 12–15% shorter paths, reduces collisions to a rate of 3.5%, and reduces overall cost to an average of 0.95. Additionally, compared to conventional baselines, the COGNIRobots framework achieved a significantly lower average path variance (0.21) and average path delay (0.95 s), demonstrating its resilience in various scenarios. The outcomes demonstrated the potential of integrating memory-based adaptation, multimodality integration, and intent-aware coordination to facilitate multi-robot navigation in realistic environments.