Digital twin technology and real-time data integration for military autonomous vehicles

The aim of the study is to detect obstacles occurring in road conditions by developing a camera and deep learning-based system and transfer this data to the digital twin platform using a 5G infrastructure. First of all, electronic circuit, PCB designs, mechanical designs and skeletal structure were developed for the vehicle in the study. A control infrastructure based on STM32-based microcontroller and minicomputer was created. In order to detect obstacles on the road, the YOLOv10 model was trained with a data set consisting of 2.000 images. The mAP value, which reached %98-%99 as a result of the training, shows that the model detects objects with very high performance. The detected objects and sensor data received from the vehicle were sent to the cloud server using the 4G infrastructure and to the end server using the 5G infrastructure. While a delay of 106 ms occurred during data transfer when 4G and cloud server were used, the delay was reduced to 23 ms when 5G and end server were used, allowing the data to be transferred and displayed on the digital twin almost instantly. Unity was used to create the digital twin. In addition, lane tracking was needed for the vehicle to proceed on the specified track while detecting objects. At this stage, a lane was created using training cones. A regression-based equation was developed according to the arrangement of these cones and their positions on the screen.