Optimizing Visual Instruction Detection in Autonomous Mobile Robots Using Yolov8 and Tensorrt Acceleration

In this study, the deep learning-based detection performance of instructions for the vehicle was examined through images obtained from a camera mounted on a mobile robotic system. The aim is to enhance the detection performance of a differential robot equipped with a robotic arm in recognizing various visual instructions it may encounter in the field. Traffic lights, direction signs, and speed limit signs were selected as the visual materials to be introduced to the robotic system. By utilizing the YOLOv8 object detection model on the embedded AI computer onboard the vehicle and leveraging the TensorRT accelerator, deep learning-based image processing achieved a high frame rate of 33 FPS and an mAP50 accuracy of 96.6%. This study highlights the advantages and challenges of integrating advanced detection models into autonomous robotic platforms, contributing to future improvements in reliability and efficiency.