Perception Sensor Simulation and Model Validation


The development of automated driving systems is becoming increasingly virtualized with the help of simulated vehicle environments. Perception sensors, as the interface between the vehicle and the open world, however constitute one of the major challenges. Persival will support the utilization of sensor models for your inidvidual use-case, from defining requirements, building custom sensor models, all the way to validiation with real world measurements. As perception sensors are also more and more considered in other industries, we are working on simulating camera, lidar and radar for robotics and infrastructure applications, too.

Sensor Model Requirements

Knowing what you need for your specific use-case is the first major step in building a simulation pipeline. Next to interfaces, hardware and parameters, selecting sensor effects that have to be modeled is most challenging.
We established the PerCollECT initiative to collect and provide perception sensor cause effect chains in a tree-shaped ontology. Utilizing our Cause, Effect, and Phenomenon Relevance Analysis (CEPRA) we specify sensor models tailored to your use-case with a traceable decision making process.

Custom Sensor Models

Environment perception sensors deal with the whole world. They are extremely complex. An off-the-shelf generic model with a hand full of parameters will not be able to handle every unique application. We developed a set of Open Source base models for active perception sensors with widely used standards like FMI and OSI. Starting from there, Persival will develop custom sensor models specifically tailored to your specific hardware and simulation needs.

Spray Simulation
Rain and Spray
Spray Simulation
Clear Weather

Sensor Model Calibration & Validation

No model will be useful, if you do not know to what extent it represents the real world system. We use sample validation methods to compare simulation results with real world measurements. Interpolation and extrapolation of metric results at these samples yields an uncertainty quantification of the model for every operation point of the model. So at every point of your simulation, you know exactly, how uncertain the model is. This methodology can also be applied to model calibration, as it allows the multidimensional comparison of the simulation model with the real system.

Model Calibration
Uncalibrated Model
Model Calibration
Calibrated Model

Model Application for Perception Development

By applying perception sensor models in simulation based testing, you can analyze your needs before you even buy any sensors. Persival will simulate different sensor systems comprised of one or multiple sensors for your specific use-case, in order to derive a specification for the real system. Use-cases can range from automotive applications, over robotics to infrastructure. With this simulation based approach, we develop algorithms for any use-case in environment perception across all industries that utilize cameras, radars or lidars. The algorithms are implemented and tested very efficiently with an agile response to customer needs before buying expensive hardware.

About Us

After studying mechatronics engineering at Technical University of Darmstadt, we worked as research associates and PhD candidates at the Institute of Automotive Engineering (FZD) at TU Darmstadt, under the lead of Prof. Dr. rer. nat. Hermann Winner.
Together with all major companies and research institues of the German automotive industry, we participated in the research projects ENABLE-S3, PEGASUS, SET Level, and VVM on simulation based development and validation of automated driving functions. We have several years of experience in the definition of sensor model requirements, model development and validation. Furthermore, we are a founding member and part of the Change Control Board of the standard for simulation model interfaces called OpenSimulationInterface (OSI) at ASAM.