What non-automotive Machine Learning projects can learn from automotive Machine Learning projects

Could your recommender system be considered high-risk AI? The EU AI Act will soon require automotive-grade safety practices for many common applications.

#1about 5 minutes

The business case for investing in AI and machine learning

Market projections, corporate investment trends, and performance data from high-performing companies demonstrate the growing value and profitability of AI.

#2about 5 minutes

Navigating the upcoming EU AI Act for high-risk systems

The European Union's AI Act classifies applications based on risk, requiring robust development practices for systems that can cause physical or mental harm.

#3about 4 minutes

How CARIAD is tackling major automotive industry shifts

CARIAD, a VW Group subsidiary, focuses on software and connectivity to address key industry challenges like electrification, digital user experience, and autonomous driving.

#4about 4 minutes

Decoupling hardware and software with a unified platform

The traditional complex vehicle architecture with over 100 ECUs is being replaced by a unified platform (SSP) and operating system (VW.OS) to enable agility and over-the-air updates.

#5about 6 minutes

Building an AI-ready architecture for autonomous driving

The goal is to transform vehicles into AI platforms by creating a virtuous cycle of data collection and model improvement, supported by powerful hardware and a connected backend.

#6about 5 minutes

Adapting the traditional V-model for ML development

The rigid automotive V-model is adapted for machine learning by splitting the software phase into distinct data, model training, and porting stages to incorporate necessary iteration.

#7about 5 minutes

Treating data as a product using the three V-model

The three V-model approach treats the dataset as a distinct product with its own requirements and KPIs, enabling better traceability and versioning across data, models, and code.

#8about 4 minutes

Advanced methods for robust computer vision models

Techniques like AugMix for data augmentation, out-of-distribution sampling for unknowns, and Delta Learning for model adaptation are used to build more robust and reliable computer vision systems.

#9about 2 minutes

Tooling principles for safety-critical ML development

The development process relies on a toolchain where everything is treated as code—including requirements, data, and models—to ensure versioning, reusability, and compliance with certified tools.

#10about 7 minutes

Q&A on data privacy, e-mobility, and managing AI hype

The Q&A covers key topics including data anonymization for privacy, the company's commitment to net-zero emissions beyond e-mobility, and the strategy for building trust by not over-promising on AI capabilities.