Artificial intelligence enters ADAS

Artificial intelligence enters ADAS

Technology News |
By Christoph Hammerschmidt

In the fourth round of the Proreta research project, the partners Continental and TU Darmstadt developed a machine-learning vehicle system that supports drivers in inner-city traffic situations and implemented a prototype. Data from radar sensors help to assess the traffic situation when turning left, entering a roundabout or passing right-over-left junctions. Machine Learning mechanisms were used in the project to generate algorithms that use various vehicle data to generate a current driving type profile of the person behind the steering wheel. On this basis, the City Assistant System’s recommendations for driving manoeuvres are adapted to the driver’s driving style.

The task of the Proreta 4 project was to use adaptive systems to implement solutions that so far could not been tackled due to a lack of adaptability. “In order for an assistance system to make a recommendation to the driver in a complex traffic situation that is accepted with confidence, the system must analyze the driver’s driving style and thus his subjective perception of safety or risk,” explains Hermann Winner, Head of the Vehicle Technology Department at Darmstadt Technical University. The scientist assumes that such a driving profile can be created comparatively safely and quickly on the basis of a machine learning process. For this purpose, the system evaluates data that is captured while the vehicle is in motion. Among other things, acceleration, yaw rates, braking processes and lateral acceleration provide the algorithm with information about the type of driver involved.

Test drives showed that the algorithms used in the City Assistant System allow conclusions to be drawn about the driver’s current driving style within three to five driving maneuvers. This makes it possible to assign the driver to one or more clusters of driving profiles, allowing the driving recommendations of the City Assistant to be strongly personalized.

Machine-learned algorithms are increasingly finding their way into vehicle systems. It is estimated that by 2015 the number of vehicle system units using artificial intelligence will increase from seven million to 225 million by 2025. Powerful machine-learned algorithms are usually highly complex models whose raw form can only be interpreted by humans to a limited extent or not at all, similar to a black box. This poses special challenges for the safeguarding of assistance systems. For this reason, a hedging strategy was co-developed as part of the algorithm selection for driver assistance systems. In Proreta 4, various methods for reducing the required number of test cases for learned algorithms were developed, which are now being researched further.

“The driver should be able to develop confidence in the City Assistant System and its recommendations,” comments Ralph Lauxmann, Head of Systems & Technology in the Chassis & Safety Division at Continental. “Trust is the basis for the acceptance of assistance systems, which in turn are an essential component on the road to accident-free driving.

On the basis of the driving profile, the system controls the time windows for driving recommendations, for example for the left-turn assistant. The system uses its own position data and the speed and distance of oncoming traffic to determine how large the gaps are in oncoming traffic for a left-hand turn. Object detection is performed by a long-range radar ready for series production and short-range radars in the vehicle sides, which are already in use in many assistance systems today, such as intelligent Adaptive Cruise Control or blind spot monitoring.

The driver needs assistance when the time window necessary for safe turning is critical or when it becomes difficult for the driver to estimate it. This can be the case at night or in poor visibility, or for inexperienced or elderly drivers. In heavy traffic, the City Assistant System reduces the stress of gap-finding and informs the driver when a suitable gap occurs. Trial runs as part of the research project determined a time window of five to seven seconds in which the system can provide assistance with recommendations. The lower value with smaller gaps in oncoming traffic applies to slightly more dynamic drivers, the upper value for very defensive drivers. The same principle applies to another application: entering a roundabout.

The driving recommendation can be given on different routes. The system can give its advice with optical, acoustic or haptic signals, depending on the driver’s preference. In general, the test drivers found an optical display with a large green or red arrow to be the best. 

In order to enhance the safety of the driving recommendations, the position of the vehicle was also included in the decision-making process. For this purpose, a camera-based system recorded prominent points in the surroundings and compared them with the navigation data. With this long-term SLAM procedure (Simultaneous Localization and Mapping), landmarks are recognized, evaluated and stored in a data memory in the vehicle on more frequently travelled routes. This makes it possible to detect positions on these routes with an accuracy of less than one meter.

Related articles:

Continental shows system to fight driver distraction

Brainstorm: Report on Artificial Intelligence

Virtualization on Microcontrollers

Autonomous delivery vehicle automates the “last mile”

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles