Test Bed Lower Saxony – Phase 2
From a scientific perspective, the greatest challenges in the field of automated and connected driving (AVD) are currently the verification and validation - i.e. testing - of the systems. Automated driving requires various sensors to detect the environment and determine position. The more accurately a sensor perceives the environment, the lower the risk of objects remaining unrecognised. With the current state of technology, for example, very high colour depths and high resolutions are already available for cameras.
This is undoubtedly impressive. However, this example also points in the direction of a central problem area: a current camera can mathematically return around 50 million different values (at Full HD resolution and 24-bit colour depth). The images captured by such a camera depict a variety of traffic situations characterised by normative and non-normative behaviour on the part of road users. These must be reliably recognised by algorithms and understood to a certain extent in order to provide a basis for appropriate vehicle behaviour. Even when reduced to a selection of relevant traffic situations, testing with conventional methods is only possible with considerable effort. In addition, for ethical reasons it is not advisable to accept critical and especially supercritical situations during testing.
Only simulation-based testing will be able to provide sufficiently comprehensive information on the safety of automated and connected vehicles.
Phase 1 of the Test Bed Lower Saxony, which will be operational in January 2020, is an essential prerequisite for the development of powerful, model- and simulation-based tool chains. In the second phase of the Lower Saxony test field, a digital twin of the test field, i.e. the environment of automated and connected vehicles and the traffic situation, is therefore to be set up at a high level of detail based on this. These are the core elements of a new large-scale digital research facility of the German Aerospace Centre.
The detailed recording of reality with the possibilities from set-up phase 1 makes it possible for the first time to create a comprehensive and validated simulation platform that can be made available to industry and research participants. We are also developing methods and technologies that will enable us to make other test fields (Pikes Peak, Highway 1, Test Field A9, TAF BW, TAF HH, AIM etc.) virtually available for the Test Bed Lower Saxony.
Standardisation projects for architectures and interfaces for simulation-based tools that are currently being shaped by us will be taken up in the implementation of phase 2. This will ensure integration capability with tool chains of industry and research participants (see, for example, standardisation activities coordinated by ASAM for openDriVE, openScenario and openCRG).
The Test Bed Lower Saxony is divided into several modules. Modules 0, 1 and 4 are currently under construction or already in operation (Module 0 - Intelligent mobility application platform (AIM) has been fully operational since 2014).
As part of phase 2, modules 2 and 6 will be realised as digital infrastructures. Selected areas are measured with high precision and all relevant parts are digitally mapped in a suitable quality.
In this context, Hanover and the BAB 2 motorway between Hanover and Braunschweig will become prototypes of an AI-based digitalisation process, which is essentially composed of ground truth data from phase 1 and environmental data recorded with, for example, series sensor technology. The aim of the second phase is to keep the track-specific effort as low as possible and to efficiently apply generalised tools from the first phase of the Lower Saxony test field. Apart from a few highlights, such as the integration of the communication units from the C-Roads project on the BAB2 motorway (Module 2) and the integration of the Hannover Messe exhibition grounds and the ADAC driving safety centre (both part of Module 6), no sensor equipment will be used. The creation of a digital map must also be reduced to an acceptable level of effort. High-precision surveying of large sections of the landscape must not become a basic prerequisite for valid models or simulations. On the contrary, only if deviations between reality and its digital image have no critical and only very predictable effects on the behaviour of automated vehicles in simulations can sufficient confidence be gained in simulation-based tested automation functions.
