For the first time, ADAS sensors have been tested in artificially reproduced rain at the Digitrans proving ground for automated driving using the facility’s new outdoor rain plant, which went into operation in late 2022. The analyses were performed as part of the InVade (Integrated Vehicle-in-the-Loop for Automated Driving and E-mobility) project being undertaken by the Institute of Automotive Engineering at Graz University of Technology, together with KS Engineering, Magna Steyr, the Institute of Vehicle Safety at Graz University of Technology and IPG Automotive.
To determine the performance of laser-based ground truth referencing and to measure the performance of radar, lidar and camera sensor technology in different weather conditions and precipitation intensities, in daylight and at night, the team performed around 306 assessments. The results will be applied to develop virtual sensor models to enrich simulation and test bench analysis.
Tests were carried out on three lanes, with four test vehicles simultaneously. They were done both during the day and at night with artificial light. The scenario selection was based on covering as many effects of the ADAS sensor technology as possible, including those that could trigger malfunctions, such as object separation and lane detection. In several test scenarios, different cut-in maneuvers were performed with multiple vehicles. Detection of the environment and moving objects by the ADAS sensors was compared with reference data of the environment and the executed test scenario.
The detection of the lane markings by the ADAS sensors as a function of the rain intensity and the quality of the lane markings was also checked during the driving tests.
Measurement technology and equipment used in the ego test vehicle:
- Dewetron CAPS (combined RTK-GPS and IMU measurement of the dynamic vehicle conditions and the driven trajectory);
- Robosense ground truth measurement system based on a high-resolution lidar;
- Various automotive ADAS sensors allowing access to object lists and partially raw data:
– Two Continental 3 series radar sensors (combined long- and short-range radar)
– Two Continental 4 series radar sensors (combined long-range and short-range radar)
– Robosense 16-beam laser scanner
– Mobileye video camera
– Cohda MK-4 Car2Car communication
- dSpace datalogger with high recording rate.
Experiences and output of the first ADAS sensor tests in the artificial rain:
Determination of precise measurement data based on the sensor tests
The 306 driving tests (of which 278 were evaluable) provide the basis for developing sensor models. The goal of sensor model development at the Institute of Automotive Engineering at Graz University of Technology is phenomenological and adopts data-driven modeling approaches, which is why a large amount of precise measurement data is necessary.
Evaluation of the predictive capability of simulated sensor models as a further step
To prove the quality of the sensor models or compare them with other approaches, the DGT-SMV method was developed at the Institute of Vehicle Technology at Graz University of Technology (technical paper on the method available here: www.mdpi.com/1996-1073/15/7/2545). This method will be further developed as the InVade program continues. It is an exact re-simulation of the driving tests performed in IPG CarMaker. With a statistical comparison of the measured data of the real and simulated sensor, the predictive capability of sensor models can be evaluated.
“The unique setup of the outdoor sprinkler system at the Digitrans proving ground in St. Valentin enables systematic and reproducible testing of ADAS sensors in rain under real conditions for the first time,” said Dr Arno Eichberger, head of the vehicle dynamics research area in the Institute of Automotive Engineering at Graz University of Technology.
Further fields of application of the outdoor sprinkler system for sensor testing in rain:
- Performance evaluation of environment awareness systems (lidar/radar/camera);
- Performance evaluation of lane marking systems;
- Data generation for environment sensor models for virtual ADAS tests;
- Data generation for benchmarks of environment awareness systems.
For more on safety, please click here.
Videos: TU Graz