Project

Our approach

"Currently, the development of AI methods for automated driving is mainly data-based. Large amounts of data are needed for training and validation. Furthermore, data-based AI functions are often black box models, their decision making cannot always be retraced. In the project KI Wissen, we address these challenges and develop methods for integrating existing knowledge in order to make the training process more efficient and to significantly increase the comprehensibility of AI decisions."

Dr. Jörg Dietrich (project coordinator)

Suppliers

Technology providers

Research facilities

Universities

Federal research institutions

External partners

Project structure

Sub project 1

Knowledge integration

Knowledge relevant to the traffic context is identified and systematized. Subsequently, the integration of knowledge in training and development of new architectures integrating knowledge for applications in autonomous driving are realized.

Sub project 2

Knowledge extraction

Different methods for extracting knowledge are examined, adapted and (further) developed. These include: methods for extracting concepts from the model, as well as models that directly provide a structured output.

Sub project 3

Knowledge conformity

Development of methods that support the plausibility check and validation of AI functions in autonomous vehicles by using different types of knowledge. The decisions of the AI functions are checked for their conformity in relation to formalized knowledge.

Sub project 4

Enabler, Integration and Demonstration

The functions, components and methods developed in the project are evaluated and demonstrated with an autonomous driving simulator using a catalog of traffic scenes.

Facts and Figures

Simon Heinz, Continental AG

Project coordinator

36 Months

Duration 01.01.2021 - 31.12.2023

€25.9 M

Budget

€17.4 M

Funding

15 partners

Project consortium