Scholarlink

<p>The success of service robots in human-centric environments relies on their ability to navigate flexibly and robustly. However, the inherent stochasticity and dynamism of human behavior pose significant challenges for robots, particularly in crowded environments. To address this issue, we introduce a deep reinforcement learning approach (ARSA), equipped with memory-assisted capabilities . This framework incorporates bidirectional gated recurrent unit layers as a long-term memory component to capture and retain the ever-shifting environment. The proposed approach emphasizes human-robot interactions by encoding their significance in decision-making. Furthermore, the learned policy incorporates the concept of dynamic warning zones to prioritize human behaviors and proactive robot decisions. The proposed model enables the robot to make better-informed decisions by identifying and focusing on the most relevant human states. The simulation results demonstrate that the ARSA policy improves the success rate by 4%, maintains the collision rate below 4%, and reduces navigation time by approximately 14%, highlighting its superior efficiency and safety compared to state-of-the-art methods. , The performance was successfully validated through real-world experiments, demonstrating smooth, collision-free ARSA behavior in the presence of dynamic human movements.</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

Autonomous robots with socially-aware navigation using memory-assisted deep reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月12日

查看全文
  • Estrella Montero,
  • Nabih Pico,
  • Manuel S. Alvarez-Alvarado,
  • José Miguel Erazo Ayón,
  • Pengkai Wang

摘要

The success of service robots in human-centric environments relies on their ability to navigate flexibly and robustly. However, the inherent stochasticity and dynamism of human behavior pose significant challenges for robots, particularly in crowded environments. To address this issue, we introduce a deep reinforcement learning approach (ARSA), equipped with memory-assisted capabilities . This framework incorporates bidirectional gated recurrent unit layers as a long-term memory component to capture and retain the ever-shifting environment. The proposed approach emphasizes human-robot interactions by encoding their significance in decision-making. Furthermore, the learned policy incorporates the concept of dynamic warning zones to prioritize human behaviors and proactive robot decisions. The proposed model enables the robot to make better-informed decisions by identifying and focusing on the most relevant human states. The simulation results demonstrate that the ARSA policy improves the success rate by 4%, maintains the collision rate below 4%, and reduces navigation time by approximately 14%, highlighting its superior efficiency and safety compared to state-of-the-art methods. , The performance was successfully validated through real-world experiments, demonstrating smooth, collision-free ARSA behavior in the presence of dynamic human movements.