Virtual tests are used increasingly to assess automated driving. They are generally based on ground truth information. The perception process of automated vehicles with sensors provides a shifted, erroneous view of the environment. Within the virtual assessment process of automated driving, it is necessary to restore this sensor view for the automated driving function to ensure realistic testing. This paper presents two approaches to derive models that are able to recreate these perception errors. Both approaches are evaluated regarding their overall performance to emulate the sensor view of automated vehicles. They are able to create models that significantly reduce the difference between ground truth and an automated vehicle's sensor view of the environment. The resulting models can be used to improve the virtual assessment process.
Comparing a Linear Filter with a ML-based Approach for Modeling Perception Errors of Automated Vehicles
2022-03-07
1260367 byte
Conference paper
Electronic Resource
English