A Hybrid Approach to Real-Time Robotic Visual Navigation: Integrating Detection and Scene Segmentation

Robotic vision-based navigation is crucial for enabling autonomous operation and task execution in various environments. However, most current research focuses heavily on achieving high accuracy, often at the expense of real-time performance and efficiency. In this study, we propose a novel approach that balances both efficiency and accuracy by integrating a semantic segmentation branch into the YOLOv5 architecture. We designed a semantic decoder dedicated to semantic segmentation that can better integrate the information of multiple layers of backbone. Our enhanced model is trained on a custom dataset extracted from COCO which is specifically designed for robotic navigation. This integration significantly improves the efficiency and speed of the system, making it suitable for real-time applications. Experimental results demonstrate the effectiveness of our approach, showcasing superior performance in real-world robotic navigation tasks.