A Framework for Simulative Evaluation and Optimization of Point Cloud-Based Automotive Sensor Sets

Sensor perception is an essential task in the field of automated driving. Accordingly, the sensor set design is an important basis for well-functioning perceptive functions. When time and cost factors are taken into account, testing and validation on real test vehicles are only possible for a few different sensor set variants. For this reason, virtual simulation is becoming increasingly relevant for the validation and design of sensor sets in an early stage of the vehicle development process. In simulative environments, different sensor sets can be compared and validated virtually at low cost concerning their sensor parameters. With this paper, we present a methodology and framework for sensor parameter evaluation and optimization using virtual simulation for point cloud-based sensors. In a proof of concept, we exemplary demonstrate an automatic optimization of Lidar sensor mounting positions using the simulator CARLA. Compared to a baseline sensor set, our optimized sensor set achieves a significant performance improvement when considering our proposed raw sensor data evaluation metric, as well as an improvement when using a benchmark 3D object detector.