Perception is the detection of static and dynamic elements of the traffic environment and high-precision self-localization within a digital map. The functional requirements are implemented by a suitable sensor setup and downstream perception algorithms in the ODD.
Main topics
Environment detection/sensor technology/integration:
A sensor setup aligned with ADS feature and ODD together with compute hardware and a software stack is put into operation in the vehicle. The system includes the synchronization, fusion and interpretation of the raw sensor data into a consistent representation of the traffic environment. This is based on a comprehensive overall solution that is tailored to the specific application.
In parallel, innovative multimodal fusion processes are being developed in order to keep pace with the rapid advances in technology. The modalities, several instances of camera, LIDAR and RADAR, are not considered individually, but rather together.
Calibration on-line and off-line:
Perception requires precise calibration of the sensors, i.e. the exact determination of the spatial poses (extrinsics) and distortions (intrinsics). In state-of-the-art systems, the calibration must first be determined off-line, i.e. once at the start of commissioning, in a workshop or special calibration environment and, if necessary, checked regularly. (Self-)calibration at runtime (on-line) is desirable. New procedures are being developed for this purpose. Another goal is to find fusion algorithms that are robust against calibration errors.
Map & localization:
The use of a digital map allows the inclusion of a priori knowledge within the perception algorithmic and represents a further source of information, i.e. a fourth modality that increases the overall availability of the system.
Provision of a public dataset
Developments in perception are significantly influenced by publicly available data sets, which are used to develop ever-improving methods. To date, these data sets have all been derived from measurements on passenger cars. For the first time, we are making a public dataset available to the research and development community, specifically for the truck sector with its special features.
