With rapid strides, Bosch Mobility, the largest corporate division of the Stuttgart-based Bosch Group, together with software specialist and VW subsidiary Cariad, is advancing the perfection of automated driving functions. Tested and systematically trained with large data volumes, the AI-controlled driver assistance solutions are now being tested on electric VW vans. In about a year, the solution for the “Software-defined Vehicle” should be ready for serial production at the Wolfsburg automotive maker.
The extended electronic control is the result of a deepened collaboration within the Automated Driving Alliance, which the two companies founded in 2022. The purpose of this cooperation is to develop Level 2 systems with the possibility of hands-free driving in certain areas of the city, on rural roads and highways using AI technologies. Another field of work are developments at Level 3 for autonomous operation of the system on highways. Test vehicles are on the roads in Europe, the United States and Japan.
Keeping Cars Up-to-Date Longer
Already a year ago Bosch announced at its Bosch Tech Day 2024 in Renningen, west of Stuttgart, that with software it still aims to generate “billions in revenue” by the end of this decade — in view of the “era of the software-defined vehicle” (Bosch managing director Dr. Markus Heyn). Bosch says it employs 48,000 people in software development – alone in the Mobility business area there should be 42,000. Heyn prophesied:
“Autos werden künftig nahtlos in die digitale Welt integriert und dadurch vor allem eins sein: updatefähig. Mit Bosch-Technik altern Autos langsamer.”
More Safety for All
The current software stack is used for the assisted and automated driving of Levels 2 and 3. The aim is to replace these driving levels completely with artificial intelligence, according to Bosch in its press release. To this end, existing approaches will be expanded with the most modern AI methods. Bosch:
„Das soll zu leistungsstärkeren und intelligenteren Fahrerassistenzsystemen führen, die so natürlich wie ein Autofahrer agieren – und damit das Fahrerlebnis auf ein neues Niveau heben und noch sicherer machen.“
With the software package, the providers promise to take over all essential rational functions of humans such as interpretation of what has been perceived, decision-making and action. The goal is to make automated driving not only accessible for shuttle purposes in public transport or transport services — but also for projects like VW’s, for example with cooperation partner Mobileye, or its Hamburg-based Moia has been involved for some time – but for all private drivers: “from the volume segment to the premium segment.” This would allow the driver,
in some driving situations, to let go of the steering wheel and be free—for example for entertainment.
There are already implementations in test fleets. These would now be “systematically trained and further developed with large data volumes,” Bosch said. They expect that by mid-2026 a software stack will be available for adoption into serial projects.
From Driver Assistance to Steering-Wheel-Free Autonomy
Many vehicles already include Level 1 of the five-level automated driving system: here the driver still controls and is responsible for operation and is only supported by certain driver-assistance systems, for example a cruise control to maintain speed or lane-keeping assistants, which can also be switched off. This level is also called assisted driving.
The semi-automated driving of Level 2 includes Advanced Driver Assistance Systems (ADAS), which means that two assistance systems are combined, for example lane-keeping assistant and distance control. In this situation the vehicle can manage some tasks autonomously without human input, for example automatic parking. There are variants. The Friedrichshafen-based supplier and mobility expert ZF, for example, also offers Level 2+-systems in which different driver assistance systems are “intelligently linked” together, as they say. Even though this level allows temporarily releasing the steering wheel — the responsibility remains with the driver, who must be ready to act if necessary. ZF reports on its website about the topic: “Texting while driving at the wheel remains taboo.”
Level 3: The Next Stage
With the next Level 3 the highly automated driving is already reached: with automatic control and steering in traffic, the “drivers” can under certain circumstances also allow themselves to read a newspaper, since constant monitoring of the traffic situation is no longer required. The driver only has to intervene if the system signals a problem. Currently, the few vehicles of this type — for example from Mercedes-Benz — may only drive up to 60 kilometers per hour in this mode. And the driver can freely use his or her position for entertainment: movies, video meetings, mobile phone use — all allowed. Only sleeping is not allowed yet and the driver must still be able to intervene after a pre-warning time if necessary.
Only with fully automated Level 4 driving can the driver at least in certain situations completely hand over responsibility to the vehicle, which then drives “completely autonomously.” However, this state applies only to defined routes, driving on highways or in parking garages, as ZF explains. After a fully automated phase the driver can again assume control. Tests for this level have also been running for some time. Germany has,
according to ZF, as the first country worldwide created the legal framework for vehicles of this level “in defined operating areas in public road traffic in normal operation.” In the field of logistics, for example, truck-maker MAN with cooperation partners in the USA is working on autonomous trucks that perform driverless transport services — in delineated logistics areas.
With Level 5 autonomous driving there is then no hold. The vehicle acts completely autonomously in all traffic situations and human intervention is no longer provided. And thus pedals or steering wheel also disappear — the human is only a passenger.
Breaking Old Mindsets
The solutions from Bosch and Cariad are intended to advance VW’s “Software-defined Vehicle” architecture. To ensure broad impact, Bosch also wants to make the technology accessible worldwide to other manufacturers as a scalable solution. This should also strengthen Germany’s position as an economic hub and stand as proof of Europe’s power position in digital solutions, says Peter Bosch, CEO of Cariad:
„We show that the German automotive industry itself masters the key technologies of artificial intelligence and automated driving. With the expertise of our developers and engineers, we secure an integral part of Europe’s digital sovereignty.“
The primary goal is automated driving systems for as many people as possible – “so that they can gain valuable time in their car again.” Mathias Pillin, CTO at Bosch Mobility, says data-driven AI models are indispensable for spreading automated driving systems on a large scale. To this end, one must “break old mindsets” – and the Automated Driving Alliance demonstrates how such concepts can be successfully implemented.
AI: Doping for the Software
Initially used only for object recognition in the perception area, AI is now active along the entire software technology chain, meaning the system combines various technical elements such as cameras and radars, helps with decision-making and implements “the safe automated control of drive, steering and braking.” Without AI, these functions would no longer be conceivable. The provider writes:
„Die automatisierten Fahrfunktionen basieren zukünftig auf einer durchgängigen, sogenannten End-to-End-KI-Architektur, in der alle Module durch den Einsatz von KI nochmals leistungsfähiger und intelligenter werden.“
They rely on cutting-edge technology from trainable, generative AI applications, which, by analyzing a variety of scenarios, deduce the probable behavior of road users. Bosch and Cariad offer a corresponding and modern development environment and a comprehensive hardware strategy that enables them to be used in all vehicle classes—scalable and future-proof.
High Data Protection, Transparency
As the basis of the developments, there is its own source code, which is fully technically controlled
and unites several functions—from standardized data protection to security, driving safety—and also transparency, it says. Also, through optimization of the source code, innovations could be produced quickly and delivered to customers, Bosch describes the development—and the overview remains preserved:
„Die Entwickler gestalten die Architektur so, dass Entscheidungen und Handlungen der Künstlichen Intelligenz sicher, nachvollziehbar und erklärbar bleiben.“
Imitating Humans
Eine weitere Funktion des Software-Stacks besteht darin, als Basis für die Integration multimodaler KI-Ansätze zu dienen: Dabei werden visuelle und sprachliche Informationen miteinander verknüpft. Man nutzt Vision Language Action (VLA)-Ansätze, die „logisches Denken und Handeln von Menschen imitieren“ können: „Ein solcher Einsatz ermöglicht ein noch effizienteres Training und noch tieferes Verständnis komplexer Verkehrssituationen“, meint Bosch. So könnten mit Hilfe von VLA „versteckte Risiken während der Fahrt“ detektiert und eine passende Reaktion ausgelöst werden.
Große Testflotte
Bosch geht davon aus, dass der entwickelte KI-Stack das automatisierte Level 2- und Level 3-Fahren auf eine sicherere Grundlage stellen kann. Bis zur Serieneinführung, die für Mitte 2026 angekündigt wird, will man nun zur Optimierung kontinuierlich „enorme Datenmengen“ sammeln, um das System noch sicherer und zuverlässiger zu machen. Dies kann nur mit einer großen Test- und Validierungsflotte sichergestellt werden, die sich im öffentlichen Straßenverkehr bewegt. Hierfür arbeiten die Teams von Cariad und Bosch auf entsprechenden Verkehrswegen in Europa, Japan und den USA – schließlich sollen die Software-Stacks kompatibel sein für den weltweiten, automatisierten Einsatz. Wie das in etwa abläuft, schildert Bosch mit folgenden Worten: „Die Entwicklung erfolgt datengetrieben, wodurch die Software mehrfach täglich mit neuen Updates und Optimierungen am Quellcode in die Testfahrzeuge eingespielt werden kann.“
Derzeit wird die Anwendung in Testfahrzeugen wie dem ID.Buzz und dem Audi Q8 eingesetzt. Und die Flotte wird zügig aufgestockt: „allein in diesem Jahr werden zusätzliche Testfahrzeuge im dreistelligen Bereich mit einem umfassenden Sensornetz ausgestattet, um qualitativ hochwertige Daten zu sammeln“, kündigt die Meldung an. Auf dieser Grundlage sollen die KI-Stacks dann weiter optimiert und besonders heikle Fahrsituationen analysiert werden, auch „Corner Cases“ genannt. Der Anbieter verspricht sich von der KI-gestützten Softwarelösung ein noch sichereres und komfortableres, „assistiertes und automatisiertes Fahren“ für den Endkunden.
Tests with Moia for years
Tests with autonomous ID.Buzz have been running for several years. The vehicles are packed with computer technology to capture the surroundings without gaps: ADAC reports in a study about the automated ID.Buzz used in cooperation with Mobileye that, for example, distances are measured with 11 radars based on electromagnetic waves as well as six lidars that operate with laser light. In addition, 14 cameras provide detection of the surroundings with objects up to 400 meters away.
