The successful conclusion of the KI Data Tooling Project unveils significant advances in Automotive AI
The highly anticipated final event of the KI Data Tooling project concluded with remarkable success, as 100 participants gathered on-site and dozens joined online to witness the unveiling of pioneering results in the realm of artificial intelligence for advanced driving technology. Hosted at the esteemed BMW Welt premises in Munich on December 5 and 6, 2023, the event showcased the collaborative efforts of 17 project partners from industry, academia, and research institutes.
Attendees were treated to a wealth of insights over the two-day event, delving into presentations, keynotes, expert talks, and poster exhibition. The first day focused on the KI Data Tooling project and the challenges addressed. The key results were wrapped into compelling hypothetical future ISO PAS 8800 audit story, taking the audience step by step through KI Data Tooling concepts and contributions for a future systematic, data-driven machine learning development approach for automated driving functions. Along with the introduction of the KI-DT Data Kit Framework designed to bolster the development of resilient automotive AI, highlights involved taking a systematic journey through the areas of
- Scenarios and Synthetic Data
- Acquisition, Curation & Refinement of Real Data
- Data Storage, Analysis & Discoverability
- and Machine Learning Function Development,
These play a crucial role in the development of AI for approved, safe automated driving functions and the associated standardization activities.
The second day provided a deep dive into specific areas, including synthetic data on the example of amodal perception with (partially) occluded pedestrians, mixed data training, real data and labeling, synthetic data production and validation, and corner cases. The event also highlighted project achievements in methods and tooling, the data-centric AI developer journey within the KI-DT Framework, and a forward-looking perspective on future developments. The culmination was marked by a panel discussion on generative AI and the regulatory challenges faced by businesses and researchers in the field.
As the project comes to a close, the consortium expressed gratitude for the dedicated collaboration that enabled the realization of a shared vision for a Data Kit for Automotive AI. Despite encountering significant risks and the need for strategic pivoting, the KI Data Tooling project concludes with contributions that surpass the current state of the art and make contributions to ongoing standardization initiatives such as ASAM OpenMATERIAL.
The journey, however, does not end here; it signifies just the beginning of a commitment to align theoretical framework architecture with practical methods. The focus now turns to meeting the demands of a robust data and machine learning training loop, with an eye on future requirements such as coming from ISO 8800.
On the KI Data Tooling Final Event the consortium invited stakeholders to join in evaluating the discussed methods and solutions, showcased real-world applications, and demonstration of the effectiveness of the developed methods. This marked the starting point on the road to credibility, and the consortium is excited to have everyone on board for this ongoing journey of innovation in AI for automated driving functions.
The outcomes of the project and materials from the concluding event will be made available to the public on this website: Results
Images ©Jürgen Biniasch