Improved Hybrid Model for Obstacle Detection and Avoidance in Robot Operating System Framework (Rapidly Exploring...

Area of Science:

• Robotics
• Machine Learning
• Computer Vision

Background:

• Real-world robot navigation faces challenges in dynamic and unpredictable environments.
• Existing hybrid methods with ROS navigation stacks struggle with real-time performance in changing conditions.
• Precision in obstacle detection and avoidance control is crucial for safe and efficient robot operation.

Purpose of the Study:

• To present a novel solution for enhancing mobile robot navigation in complex, dynamic environments.
• To improve the real-time performance and safety of robot navigation systems.
• To leverage advanced object detection and path-planning algorithms for superior navigation capabilities.

Main Methods:

• Integration of a pre-trained YOLOv7 object detection model for accurate obstacle identification.
• Combination with a rapidly exploring random tree (RRT)-integrated Robot Operating System (ROS) navigation stack.
• Utilizing the dynamic windows approach (DWA) for efficient path planning and obstacle avoidance.

Main Results:

• The NAV-YOLO system demonstrated high-level obstacle avoidance capabilities in simulations and real-world experiments.
• Improved navigation performance in complex and dynamically changing settings.
• Enhanced safety and efficiency of mobile robot operations, particularly noted in aviation environments.

Conclusions:

• The proposed approach effectively addresses the challenges of mobile robot navigation in dynamic environments.
• The integration of YOLOv7 and RRT-based ROS navigation significantly enhances robot safety and efficiency.
• This solution offers a promising advancement for industrial mobile robot applications.