A real-time simulation environment architecture for autonomous vehicle design

Purpose: This study aims to show the necessity of hardware-free simulation tools for autonomous vehicle design with a manageable production process in terms of time and hardware costs. To this end, a real-time simulation environment architecture is presented, including a set of required components and the feasibility of the proposed architecture is examined through the experiments conducted.Theory and Methods: To investigate the feasibility of the proposed real-time simulation architecture, an autonomous driving model, including lane tracking and traffic sign detection is introduced. Canny edge detection algorithm and different YOLO and R-CNN versions are deployed for lane tracking and traffic sign detection, respectively. The simulation architecture is tested with the components of the proposed autonomous driving model.Results: The proposed simulation architecture is evaluated on the accuracy rate of the object detection algorithm deployed for each simulation run. For this purpose, YOLO-v3, YOLO-v3 tiny, YOLO-v4, YOLO-v4 tiny, Fast R-CNN, Faster R-CNN, and Mask R-CNN algorithms are considered through the experiments. A street traffic sign dataset for Turkey is utilized for model training. According to the evaluation results, YOLOv4 is noted as the model that produces the highest accuracy rate with 95% throughout the evaluation.Conclusion: The evaluation results obtained from the simulation environment can be claimed as successful enough to use the proposed simulation architecture with different autonomous driving models for an autonomous vehicle design process without any real-world hardware cost.