Scholarlink

Navigation of Autonomous Mobile Robots (AMRs) in dynamic environments is challenging due to unpredictable and unmodeled obstacles that can disrupt pre-planned paths. This paper presents an affordance-based navigation approach that leverages machine learning with YOLOv8 for real-time obstacle detection and classification. The robot intelligently distinguishes between movable and non-movable obstacles, enabling context-aware interaction. By incorporating Gibson’s theory of affordances, the robot decides whether to push a movable obstacle using its robotic arm or replan its path around a non-movable one. The system operates on an embedded Raspberry Pi platform, ensuring efficient on-device processing, and utilizes IoT connectivity for remote monitoring and data analysis. Experimental evaluations in a smart manufacturing lab demonstrated a 90% navigation success rate, with YOLOv8 achieving 87% precision for movable objects and 81% for non-movable ones. The robot completed obstacle-laden navigation tasks with only a 19% increase in time compared to obstacle-free conditions, highlighting the system’s adaptability and efficiency. Future work will aim to improve object manipulation capabilities and extend the system for multi-robot collaboration in more complex environments.

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

Enabling Autonomous Navigation in Dynamic Environments: A Robot’s Perspective on Affordance-Based Mobility and Interaction with Movable Obstacles

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Amar Virupaxi Kavalapure,
  • Uma Mudenagudi,
  • Shridhar Doddamani,
  • Sachin Karadgi,
  • Amogh Sutar

摘要

Navigation of Autonomous Mobile Robots (AMRs) in dynamic environments is challenging due to unpredictable and unmodeled obstacles that can disrupt pre-planned paths. This paper presents an affordance-based navigation approach that leverages machine learning with YOLOv8 for real-time obstacle detection and classification. The robot intelligently distinguishes between movable and non-movable obstacles, enabling context-aware interaction. By incorporating Gibson’s theory of affordances, the robot decides whether to push a movable obstacle using its robotic arm or replan its path around a non-movable one. The system operates on an embedded Raspberry Pi platform, ensuring efficient on-device processing, and utilizes IoT connectivity for remote monitoring and data analysis. Experimental evaluations in a smart manufacturing lab demonstrated a 90% navigation success rate, with YOLOv8 achieving 87% precision for movable objects and 81% for non-movable ones. The robot completed obstacle-laden navigation tasks with only a 19% increase in time compared to obstacle-free conditions, highlighting the system’s adaptability and efficiency. Future work will aim to improve object manipulation capabilities and extend the system for multi-robot collaboration in more complex environments.