Shader-based sensor simulation for autonomous car testing

During autonomous car system development, sensor simulation can help to test and evaluate algorithms such as sensor fusion and object tracking in simulated dynamic scenarios at an early stage; thus, time and cost can be spared and more reliable system can be guaranteed. In this paper, shader-based LiDAR and Radar simulations are extended into autonomous car testing. Besides realizing sensor simulations producing information of interest scan data, conceptual programming interfaces to full featured physical models are also provided. Simulation accuracy is discussed and corresponding improvement methods are proposed. Optimistic results are displayed with a software-in-loop test for autonomous car and the computational cost is reported. Comparison between ray-tracing based and shader-based LiDAR simulation in terms of computational cost is also carried out and discussed.