Research And Implementation Of Complex Driving Scenario Simulation Model Based On Digital Twin

With the development of autonomous driving technology,the test scenarios which required by algorithm become more and more complex.Factors such as weather,traffic,and actual road conditions in driving scenarios have great uncertainty.It is difficult to evaluate driving algorithms with accurately and comprehensively by collecting road data.In real road conditions,there is an obvious long-tail effect in autonomous driving testing.The more completive the algorithm,the fewer scenarios that can improve the algorithm testing.Even if a complex traffic scene is built in a closed site,the cost is still high,and the testing process is also accompanied by high safety risks.To overcome the above difficulty,this thesis explores efficient and low-cost digital simulation models based on actual complex driving scenarios.Based on digital twin technology,this thesis implements a complex driving scene simulation model,constructs a digital twin environment through SUMO and r FPro simulation tools,and integrates technologies such as perception,positioning and mapping to achieve 1:1high-precision reconstruction of physical space.The vehicle terminal obtains control commands from the server in the control center.It makes corresponding actions according to the commands and uploads the positioning data to the server and performs virtual-real mapping and inference.The real scene is restored by the simulated car-following model and lane-changing model which new driving behavior commands are obtained through calculation.During the communication process,the edge vehicle side needs to upload data multiple times,which will consume a lot of bandwidth resources.It affecting the accuracy of the global driving behavior evaluation.This thesis proposes a multi-part uploading algorithm for this purpose,which realizes the clipping and partitioning of uploaded data.This way improves the speed of data uploading on the vehicle end,and accelerates the overall efficiency.The complex driving scene model system is designed and the corresponding functions are implemented.The experimental verification and analysis of the system are carried out.The main work is as follows:Based on SUMO micro-simulation framework and r FPro scene restoration tool,a complex driving scene model based on digital twin is built to restore the typical street scene driving scene.Through the driving scene model,various indicators such as speed,steering wheel angle and scene pictures can be obtained.（2）Based on Node JS,Python language and Express framework,the virtual server platform（central cloud server）of digital twin is built,and the basic functions such as positioning data mapping,heartbeat detection,and file fragment uploading are realized.（3）Study the car-following and lane change model algorithms.Compare the difference between different algorithms.Predict the trajectory of the vehicle through the neural network and predict the speed,angle and positioning parameters at the next moment,in order to make the appropriate driving behavior.（4）The functions of the system are analyzed and tested,and the commonly used model parameters and simulation results are stored in the management platform for users to find and configure the driving scene simulation model parameters.