Classification Of Sensor Errors For The Statistical Simulation Of Environmental Perception In Automated Driving Systems

A virtual world provides a completely controlled and safe testing environment for the development and testing of current and future automated driving systems. In order to provide conditions close to reality, the input data for the automated driving system generated by sensors for environmental perception have to match closely between virtual and real world. The data gathered by perception sensors like radar, lidar or camera sensors generally provide a lossy description of the environment. Therefore, a sensor error model has to be employed that captures the characteristics of the sensory perception process. We propose a general description of a statistical sensor model, constructed to achieve equivalent sensor output on a statistical level. To construct the model, we define model units that each deal with a specific aspect of the sensory perception process on the object level. We propose a classification scheme and hierarchy for the different error types and describe a methodology for using real world reference data as input for the statistical model.