Planning and scheduling for autonomous mobile robots using a deep learning method

Abstract

Object detection has become critical in real-world applications, including autonomous driving, robotics, and quality control in manufacturing. This study presents the design and evaluation of a hybrid object detection pipeline that combines the strengths of YOLOV8 and Faster R-CNN to enhance detection accuracy and confidence. The primary focus is to address the limitations of individual models by leveraging their complementary strengths: YOLOV8's real-time detection capabilities and Faster R-CNN's precision in refining low-confidence predictions. This study highlights the trade-offs between computational efficiency and detection accuracy, as lower confidence thresholds captured more regions for refinement but increased computational costs, while higher thresholds were computationally efficient but risked missing opportunities to refine uncertain detections. The hybrid pipeline's flexibility was also demonstrated, showing potential adaptability to other datasets or object detection tasks by tuning thresholds and exploring alternative backbone architectures. The proposed pipeline represents a step forward in combining lightweight real-time detectors with accurate refinement models to achieve a balance between speed and precision. Future work will focus on optimizing the pipeline for real-time deployment, expanding the dataset for improved generalization, and exploring transformer-based architectures for further performance gains. This study underscores the importance of model integration in tackling complex object detection tasks and provides a solid foundation for developing advanced hybrid systems.