Partnership Automated Driving: Bosch and Cariad are Working on an AI-Based Software Stack

Related Vendors

Robert Bosch GmbH

dataTec AG

Wuxi InfiMotion Technology Co., Ltd.

As part of their Automated Driving Alliance, Volkswagen subsidiary Cariad and automotive supplier Bosch are further developing their software stack for automated driving at SAE Level 2 and 3 through the use of artificial intelligence. To achieve this, the partners are expanding their existing approaches with new AI methods.

"Data and AI are the key when it comes to bringing automated driving systems to the road on a large scale and reliably," explains Mathias Pillin, CTO at Bosch Mobility. The Automated Driving Alliance therefore aims to rely even more extensively on artificial intelligence.

The two partners—automotive supplier Bosch and Volkswagen's software subsidiary Cariad—aim to make automated driving available for private motorists, and not just for the premium segment. With the new driving functions, the driver will be able to take their hands off the steering wheel in various driving situations in the future. The first implementations in test fleets are already in place, which are now being systematically trained and further developed daily with large amounts of data. From mid-2026, a software stack will then be available for application in series projects.

The automated driving functions are intended for the software-defined vehicle architecture within the Volkswagen Group; however, Bosch will also make the scalable solution available to other manufacturers worldwide.

AI Plays an Important Role From the Very Beginning

The companies have relied on artificial intelligence since the beginning of their collaboration, for example, for object recognition. Meanwhile, AI is being used throughout the entire software technology chain: from object recognition to decision-making, and to implementation in the automated control of drive, steering, and braking. The automated driving functions will be based on a seamless end-to-end AI architecture in the future. The focus is on technology as it is known from generative AI applications. Just as language models understand complex semantic relationships, the new AI stack of the Automated Driving Alliance can analyze urban traffic scenarios. It can anticipate the current and potential behavior of road users from different sensor modalities.

The end-to-end development of all technology elements with proprietary source code and intellectual property forms the basis of the development partnership. This enables full technical control of the source code with clear standards for data protection, security, driving safety, and transparency. Moreover, source code optimization allows for the agile and rapid generation and delivery of innovations. The developers design the architecture so that the AI's decisions and actions remain safe, traceable, and explainable.

The software stack also creates a foundation for the possible integration of multimodal AI approaches that combine visual and linguistic information. Vision-Language-Action (VLA) approaches can imitate human logical thinking and actions. Such an application helps make training more efficient and better understand complex traffic situations. VLAs can, for instance, assist in detecting hidden risks during driving and responding appropriately to them.

Validation Fleet in Public Road Traffic

According to the two partners, the AI stack is intended to make automated driving at Levels 2 and 3 even more robust. By the time of its series introduction in mid-2026, its performance is to be consistently improved through the continuous collection of enormous amounts of data. A testing and validation fleet in public road traffic is essential for this. Both companies are jointly testing the driving functions in public road traffic in Europe, Japan, and the USA. The development is data-driven, allowing the software to be updated and optimized in the source code several times a day and deployed to the test vehicles.

The technology is already being used in test vehicles such as the ID.Buzz and the Audi Q8. This year alone, additional test vehicles in the triple digits will be equipped with a comprehensive sensor set to collect high-quality data. This data is used to further optimize the AI stack and analyze complex and rare driving situations. (se)

Secure and Compliant Authentication in Laboratories

Secure & Contactless User Authentication at GMP/GLP