AVL Dynamic Ground Truth System™ - Engineering Services
Capture true reality
AVL Dynamic Ground Truth System™
Highly accurate dynamic environmental measurement for ADAS/AD homologation
In the development of ADAS/AD systems, how can we trust that what the vehicle sees with its arrays of sensors reflects an accurate and safe interpretation of the real world? How can we be sure that there are no distortions to the vehicle’s perception of its environment, and that it is operating in a true representation of reality – so called ground truth?
Having an accurate benchmark of the real world lets you measure the accuracy of the vehicle systems – and therefore engineer their safety. The closer the vehicle’s perception is to ground truth, the more accurate its picture of its surroundings will be. Being able to measure accuracy in this way supports development and homologation.
Accurate Representation of Reality
To support our customers in this endeavor, we have created the AVL Dynamic Ground Truth System™ (DGT system). This is a one-box solution which is mounted to the test vehicle. It lets you capture an accurate 360° field of view of a vehicle’s environment for the statistical comparison and validation of the ADAS/AD system under test (SUT). The image it captures is comprised of lidar, camera and high-precision GPS sensors, offering an extremely accurate view of the vehicle and its systems’ test environments. Optionally, radar sensors enhance the system for dedicated applications. The data streams of the DGT system and the vehicle sensors are recorded in a time-synchronized way in order to guarantee a highly precise comparison of the DGT system and the SUT. Raw and object data are visualized via an online user interface.
During test trips, petabytes of data are collected via the DGT system and the vehicle under test. These data are transferred to a data center once the vehicle returns to the garage. From there, the data are available for further processing, based on our cloud based ADAS/AD Big Data and Analytics Platform.
As an example, recorded raw data are processed by the AVL Offline Perception neural network algorithm. The outcome, a so-called object data stream, contains information about recorded objects such as cars, pedestrians, motorbikes, traffic signs, etc. during the test ride. This information is then used for activities such as comparison of SUT vs. DGT system recorded objects, and compiled into a report showing the differences. Objects can also be visualized with the Scenario Viewer. And a powerful Data Search can find specific objects, as well as raw data sections of interest.
This is what “Data-Driven Development” is all about: a statistical analysis of real-world data and a comparison of this data to the ADAS/AD functions developed for autonomous driving.
Homologation, Compliance, Quality
The ultimate target of AVL’s DGT system is to enable homologation of future AD-systems and functionalities. Furthermore, the DGT system is a powerful solution for the improvement of the efficiency of the development process for ADAS functions.
The quality of the reference data has to be significantly higher than the quality of the SUT, so that the results of the comparison – and therefore the performance of the vehicle system – can be completely trusted. This necessary condition is not only fulfilled by highly precise sensors but also with powerful offline data processing in a data center, as well as the application of methods like backward and forward tracking of the system.
With its highly flexible sensor setup, configuration and modular mounting concept it offers a great deal of flexibility for a variety of different applications for passenger cars and trucks. And with the support of our global network of highly skilled experts we can support your global data collection and provide you with service and maintenance services, assuring consistency and confidence wherever you are.