.jpg)
ADAS needs an intelligent backend
Automated driving functions and intelligent driver assistance systems (ADAS) play a crucial
role in the positioning of vehicle manufacturers (OEMs):
- Innovative, modern and cool: ADAS features position the brand
- Increased safety: ADAS features contribute to enhancing road safety, even in the
- face of considerably increased vehicle density
- More comfort: ADAS features revitalize the joy of automotive mobility
However, ADAS level 3 specifically imposes very rigorous standards for the delivery and
functionality of ADAS driving functions (see also: Kraftfahrt-Bundesamt -
Automatisierungsstufen (kba.de) (German Federal Motor Transport Authority -
Automation levels): The product must have the capability to address any anticipated or
unforeseen hazards that may arise for road users, regardless of the prevailing conditions or
circumstances.
Why is additional information so important for ADAS functions?
Given the availability and quality of service of vehicle connectivity in various environments, it is imperative for the vehicle to autonomously evaluate circumstances and take appropriate actions to mitigate potential risks. The most convenient approach would involve depending on the emergency measures provided by the on-board assistance systems. Nevertheless, this solution should only be employed as a last-minute maneuver, as it has the potential to jeopardize the flow of traffic and result in an unsettling driving experience for the vehicle’s occupants.
Hence, advanced procedures and artificial intelligence should be accessible via off-board IT in the backend to provide continuous assistance to the fleet – whether it be from up-to-date data sourced from external providers or insights gathered from the company's own fleet. This data holds significant value in aiding the on-board IT of the ADAS functions. Additionally, it is crucial to gather information on specific or unique driving scenarios to effectively train the AI within the vehicle during the development of the ADAS features.
Finally, there might be a need for specific information that is not explicitly available in the accessible sources, such as in exceptionally uncommon or ever-changing situations. Such information is essential for creating test cases and organizing test drives, as well as for preparing training data for on-board AI to recognize situations or objects.
Real-time challenges – an illustration of map data
A fundamental aspect for all ADAS driving features is an accurate map of the immediate environment for navigation. However, map information can become obsolete rapidly due to roadworks or temporary road closures.
The primary objective of map service providers is to promptly incorporate these modifications. In order to achieve this, they typically offer regular updates. Nevertheless, real-world changes occur in real-time, irrespective of the release schedules of different map service providers. Changes are implemented solely when there is an ample understanding of them, such as only following the completion of a recent 3D survey.
How is additional information collected for the ADAS functions?
It is only possible to accurately capture the extra, particularly up-to-date, data by physically driving along the specific road sections.
Data from fleet trips or dedicated individual trips can be utilized for this purpose, along with services from information providers that depend on corresponding trip data. The predicament lies in the fact that in-house dedicated fleets come with a price tag similar to that of external service providers like Google Street View. Conversely, dedicated trips are only practical in rare instances, such as for initial research or proof-of-concepts (PoC), and the specific destinations or routes for these trips must be predetermined.
Customer vehicle fleet as data collector: further challenges
Despite the fact that automotive OEMs have robust customer vehicle fleets as a viable option, this presents additional obstacles:
- The sales price and original purpose are the driving factors behind the sensor technology quality in series vehicles, resulting in certain limitations for its direct utilization in off-board data products. An illustration can be seen in the determination of location in geo-coordinates, where the position error in the vehicle data can range from a few centimeters to several meters. Moreover, this error can be further magnified by environmental factors such as tunnels and shading.
- The number of passages along road sections fluctuates significantly – both in terms of time and location – and could potentially be insufficient for specific scenarios or routes, particularly when precise, confidently measured values are necessary, such as those derived from an FMEA analysis.
- Ensuring brand integrity and meeting legal obligations necessitate strict adherence to customer consent, especially in compliance with the #DSGVO for justifying the use and processing of each unique case.
- Procedures for extracting conclusions or acknowledgments from series fleet data necessitate confirmation or validation through “ground truth”. Obtaining this dependable data on a large scale is typically challenging due to its limited availability. Video clips with geo-coordinates annotations frequently require manual viewing and labeling, which is a laborious, error-prone, and costly process.
- The customer vehicle fleet relies on secure and cost-effective back-end support, including processes, pipelines, and processing steps, to ensure the availability of up-to-date information for supporting ADAS functions at all times.
Backend optimization for vehicle fleets – msg has the expertise
msg is a well-known provider of IT services for customers in the automotive industry, including ADAS functions.
An illustration of this can be seen in a lengthy project involving a vehicle fleet: msg devised the technical architecture and design for the #Cloud backend and successfully implemented it in a manner that allows for the processing of messages in the tens of billions, with a data volume reaching well into the double-digit terabytes, on a monthly basis.
Throughout the duration of the partnership, the msg team successfully implemented a significant number of ingest components and data products, ensuring exceptional availability. Thanks to our technical expertise, we have successfully decreased continuous operational expenses by a significant amount. Furthermore, our expertise in the utilized technologies and thorough examination of potential error sources allowed us to enhance the systems in a way that they can effectively handle internal and external errors without the need for manual intervention. Thanks to the implemented resilience mechanisms, there is no longer a need for “on-call duty” outside regular office hours.
Furthermore, the msg team played a crucial role in influencing the content of the data products, and even collaborated on the initial design of some of them. Our end-to-end experience in every aspect of the cloud backend has resulted in the development of a data product that enables highly effective and worldwide identification of potential risks. This product plays a crucial role in enhancing the accessibility of an ADAS service. For another data product, we employ a highly automated MLOps mode to manage the entire AI route.
Specifically, the form of partnership – transparent, cooperative collaboration with end-to-end accountability in a DevOps model – has resulted in the quickest deployment times from conception to execution with comprehensive automation, compliance, resilience and maintainability.
