Scholarlink

<p>Traditional robot navigation had focused on avoiding obstacles, but as robots integrate into human-centric spaces, socially-aware navigation is crucial. This research paper introduces a method for navigating crowded environments while adhering to social norms, addressing key challenges like understanding human behavior, predicting movements, and making safe, efficient decisions. Unlike prior approaches, which use generic attention mechanisms, the proposed method employs a tailored attentive pooling module within a deep reinforcement learning (DRL) framework. This module specifically focuses on assigning dynamic importance scores to humans based on their interactions with the robot, prioritizing those most likely to influence its path. The interaction module uniquely models both human–robot and human–human interactions using local maps with coarse resolution, enhancing the robot’s ability to adapt to dynamic environments. The pooling module synthesizes these interactions into a fixed-length vector, which the planning module uses to evaluate the combined robot-crowd state for optimal navigation, including safe avoidance of potential collision zones. In simulations with six humans, the algorithm achieved a 98% success rate and a 2% collision rate, outperforming SARL, CADRL, and LSTM-RL. Real-world deployment demonstrated effective navigation at 0.4&#xa0;m/s, maintaining social norms. The proposed method offers a distinct and effective approach to socially-aware navigation.</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

Improvised robotic navigation using deep reinforcement learning (DRL) towards safer integration in real-time complex environments

  • 期刊论文

  • 发表日期：

    2025年1月21日

查看全文
  • Kiran Jot Singh,
  • Divneet Singh Kapoor,
  • Khushal Thakur,
  • Anshul Sharma,
  • Anand Nayyar,
  • Shubham Mahajan

摘要

Traditional robot navigation had focused on avoiding obstacles, but as robots integrate into human-centric spaces, socially-aware navigation is crucial. This research paper introduces a method for navigating crowded environments while adhering to social norms, addressing key challenges like understanding human behavior, predicting movements, and making safe, efficient decisions. Unlike prior approaches, which use generic attention mechanisms, the proposed method employs a tailored attentive pooling module within a deep reinforcement learning (DRL) framework. This module specifically focuses on assigning dynamic importance scores to humans based on their interactions with the robot, prioritizing those most likely to influence its path. The interaction module uniquely models both human–robot and human–human interactions using local maps with coarse resolution, enhancing the robot’s ability to adapt to dynamic environments. The pooling module synthesizes these interactions into a fixed-length vector, which the planning module uses to evaluate the combined robot-crowd state for optimal navigation, including safe avoidance of potential collision zones. In simulations with six humans, the algorithm achieved a 98% success rate and a 2% collision rate, outperforming SARL, CADRL, and LSTM-RL. Real-world deployment demonstrated effective navigation at 0.4 m/s, maintaining social norms. The proposed method offers a distinct and effective approach to socially-aware navigation.