Challenges and solutions
Sensors and safety sytems in adverse weather conditions
Adverse weather conditions present a major challenge for the perception systems of autonomous vehicles. Active safety systems, such as Automated Emergency Braking (AEB), must work reliably and safely in critical situations, even in rain, fog and challenging lighting conditions.
To improve the robustness and safety of ADAS/AD vehicles, AVL Software and Functions in Roding is developing the so-called “Roding Weather Dataset” (RWD). This dataset contains typical driving scenarios such as cut-in scenarios in defined, quantified weather and light conditions as well as unusual objects such as exhaust pipes and tires in these adverse weather situations.
State-of-the-art indoor laboratory for ADAS/AD sensor validation in Roding
Our data was collected at the “AVL Center for Mobility and Sensor Testing” in Roding, a state-of-the-art facility where special weather conditions can be recreated for the validation of ADAS/AD sensors. The unique indoor laboratory covers 1,600 m², including a 1,000 m² rain and fog facility. This enables precise weather simulation for certification and performance testing. An advanced water recycling system ensures environmental sustainability. The center can simulate critical lighting conditions such as dusk and dawn by continuously varying the illuminance and backlight color.
1. Sensor Setup
The data collection was realized with the AVL “AutBus”, an important part of a cooperation project for autonomous public transportation. The VW T6.1 was converted into an autonomous vehicle with the AVL “Dynamic Ground Truth (DGT)” system, which integrates all necessary sensors. Our data set includes extensive data from one camera and three LiDAR sensors.
2.1 Camera
We use a high-resolution Dalsa Genie Nano C4030 pinhole camera. The RGB images are available as PNG files.
2.2 Lidar
The LiDAR setup includes three Velodyne LiDARs: one VLS128 and two VLP16 units. Data is stitched during post-processing and provided as PCD files, containing 3D coordinates (x, y, and z) in millimeters.
2.3 Synchronization of sensors
Sensors are synchronized using GPS timestamps, operating at 10Hz. Each file is named with the UTC frame timestamp.
2.4 Sensor Calibration
Calibration is performed at the AVL Roding Calibration Center. Calibration files are available in the ‘calibrations’ folder, and projected point clouds in the ‘projected_point_cloud’ folder. Calibration is based on the VLS128 central LiDAR, followed by alignment with the rear axle center of the ego vehicle as per ISO 8855 standards.
Test data sets available
The scenarios from the table below are available for download. We also offer a free trial version. If you are interested, please send us an e-mail using the button below. We will be happy to provide you with a download link.
Main Scenarios overview
| Number | Name | Description |
|---|---|---|
| 1 | P_E-stopps | Pedestrian is placed 5m out of rain area and ego vehicle moves towards pedestrian and stopps before reaching pedestrian |
| 2 | P_E-stopps | Pedestrian with umbrella is placed 2m inside rain area and ego vehicle moves towards pedestrian and stopps before reaching pedestrian |
| 3 | T_E-stopps | Ego vehicle moves towards target vehicle and stopps behind |
| 4 | T_E-stopps_tunnel-exit | Ego vehicle stopps behind target vehicle at tunnel exit |
| 5 | T_E-avoids_tunnel-exit | Ego vehicle avoids target vehicle at tunnel exit |
| 6 | C_E-avoids | Ego vehicle avoids cyclist |
Naming Example:
T_E-stopps_tunnel-exit
T: Target vehicle
E_stopps: ego vehicle stopps behind target vehicle
tunnel_exit: light scenario as at a tunnel exit
Namings:
P Pedestrian
E Egovehicle
T Target vehicle
C Cyclist