Scholarlink

<p>This work presents a guided navigation method for unmanned ground vehicles (UGVs) that lack sophisticated navigation sensors such as LiDAR, onboard cameras, or IMUs, counting only on odometry while relying on external cameras and high-level directional commands. The proposed system employs a network of three pan-tilt cameras integrated with YOLOv8 for robot detection, a particle filter for probabilistic localization, and qualitative spatial reasoning (StarVars calculus) to generate projective directional instructions interpretable by any robot architecture. Implemented in the Gazebo simulation environment with ROS 2, the approach guides a Husky robot through environments by issuing only high-level commands such as <i>move forward</i> or <i>turn right and move forward</i>. Experimental results across 100 randomized trials demonstrated a 93% success rate in reaching target regions, despite uncertainties in initial orientation and sensor noise. Additional experiments confirmed consistent performance across repeated runs and in complete navigation tasks. These findings highlight the feasibility of adapting guided navigation methods based on relative direction reasoning to different robotic platforms, providing a resilient solution for scenarios where onboard perception is unavailable or fails. </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

Guided Navigation of Mobile Robots through High-Level Commands Using External Cameras

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月3日

查看全文
  • Gustavo Daniel Fernandes,
  • Danilo Hernani Perico

摘要

This work presents a guided navigation method for unmanned ground vehicles (UGVs) that lack sophisticated navigation sensors such as LiDAR, onboard cameras, or IMUs, counting only on odometry while relying on external cameras and high-level directional commands. The proposed system employs a network of three pan-tilt cameras integrated with YOLOv8 for robot detection, a particle filter for probabilistic localization, and qualitative spatial reasoning (StarVars calculus) to generate projective directional instructions interpretable by any robot architecture. Implemented in the Gazebo simulation environment with ROS 2, the approach guides a Husky robot through environments by issuing only high-level commands such as move forward or turn right and move forward. Experimental results across 100 randomized trials demonstrated a 93% success rate in reaching target regions, despite uncertainties in initial orientation and sensor noise. Additional experiments confirmed consistent performance across repeated runs and in complete navigation tasks. These findings highlight the feasibility of adapting guided navigation methods based on relative direction reasoning to different robotic platforms, providing a resilient solution for scenarios where onboard perception is unavailable or fails.