VIENNA, AUSTRIA – Following the successful milestones at Beko’s manufacturing facilities and Milan Bergamo Airport, the ACCOMPLISH consortium recently convened in Vienna for its third major demonstrator workshop. This session, held on April 16, focused on the Automotive sector, where the project is setting understanding how advanced AI can enhance in-vehicle cabin understanding, in a compliant way (without compromising driver privacy or without introducing data biases in AI operations).
Enhancing In-Cabin Awareness via Edge-AI
The Automotive demonstrator is dedicated to improving safety in both manual and automated driving modes. During the session, partners aligned on the technical details for integrating E3D and VIF’s driver monitoring and warning systems into real-life vehicles. A key success of the workshop was the definition of the way the Edge-AI directly could be integrated on-board in the MAN’s buses, for public transportation purposes.
To ensure high performance with low energy consumption, the system will utilize low-power hardware, such as ARM/Raspberry Pi SoCs equipped with Hailo AI accelerators, paired with RGB cameras. What was discussed was how to adapt an AI system that works perfectly in the lab, into a vehicle, uphold the project’s rigorous privacy standards (e.g. raw data will not leave the vehicle).
The workshop brought together a powerhouse of European expertise:
- VIF: Provided their research vehicle for live exploration and testing of the monitoring setup.
- Emotion 3D (E3D): Demonstrated their computer vision and machine learning setup live to showcase its full potential for in-cabin analysis.
- MAN-TR: Bringing critical safety considerations for the commercial bus and truck sector.
Tailored Safety for Trucks and Buses
A significant portion of the discussion focused on the specific needs of Original Equipment Manufacturers (OEMs) like MAN. For the bus use case, the partners decided to focus exclusively on the professional driver (e.g., detecting fatigue, distraction, and reactions to rear-view mirror events). Monitoring passengers was explicitly excluded to maintain a focus on accuracy, complexity management, and strict privacy.
Navigating the Intersection of AI and Data Operations
The workshop also served as a deep dive into the ACCOMPLISH Toolkit (WP3) with a technical overview provided by TXT. Partners mapped out how specific modules will support compliance with GDPR, the AI Act, and ASPICE standards throughout the development lifecycle. This was important to discuss as the OEM and E3D and VIF needed to know which steps of the compliance lifecycle they are currently addressing “manually” could have been done also with ACCOMPLISH. But to do that, they needed at least initial detail description of the WP3 components, in order to map their data/AI operations, with what ACCOMPLISH can provide. From WP6 perspective this was essential in order to set up a verification plan that would use dedicated KPI in order to assess the added value of ACCOMPLISH, once the alfa version will be made available for the automotive partners.
Key steps / compliance operations that were discussed pertinent and technically applicable for the Automotive Use case were:
- Data Preparation: Mapping flows from acquisition to pseudo-anonymisation and storage.
- Model Compression: Moving models efficiently the AI model from laboratory to the vehicle’s limited hardware.
- Model Performance Monitoring: Monitoring model performance and security during and after the transition to the real-life vehicle
Measuring Value and Moving to Berlin
The team worked extensively on validation tables to define the operations on data/AI and the steps necessary for an accurate “As-Is” vs. “To-Be” evaluation. Technical partners and OEMs are now finalizing these mappings to establish a clear baseline, to test when the ACCOMPLISH toolkit becomes available.
The plan for such validation will form a core part of Deliverable 6.1, the detailed execution plan for the demonstrator. All decisions made in Vienna will be consolidated and presented at the upcoming General Assembly #3 in Berlin, proving that compliance automation is not just a regulatory hurdle, but a driver of long-term business value and road safety.
