Understanding how AI works: CCAM Edition
AI has revolutionised Connected, Cooperative and Automated Mobility (CCAM) solutions by enabling AI models to be trained on vast amounts of data. However, AI remains underexplored in terms of explainability, privacy, ethics and accountability. AIthena project aims to contribute to building Explainable AI (XAI) into CCAM development and testing frameworks by exploring three main AI pillars: data management, models, and testing.
So far, the AIthena project has developed a human-centred methodology to derive trustworthy AI dimensions from user-identified group needs in CCAM applications. A set of Key Performance Indicators (KPI) for XAI has been proposed and an analysis will explore trade-offs between these dimensions. Demonstrators (use cases) will demonstrate the AIthena methodology in four critical use cases: perception, situational awareness, decision and traffic management.
As part of the AIthena project, four use cases have been established to further dissect and understand how AI can be explained and understood throughout its decision-making process. The first use case examines perception, the second examines understanding, the third examines decision making and the fourth examines traffic management. Each use case is an in-depth analysis of how different data sets are considered in the perception, understanding, decision-making and deployment of AI in vehicles.
For the first use case, data cards, sensors and visualisation tools are of paramount importance as they gather the necessary information to detect pedestrians in the traffic environment.
The second use case explores algorithms and perception data from the vehicle to explain how AI makes its decisions.
The third use case examines how AI explains its decisions to humans, or in this case, drivers in the vehicle, to build trust in the technology.
The fourth use case examines how AI-equipped vehicles are deployed in traffic environments and how these vehicles can impact or improve traffic dynamics in terms of comfort, efficiency and safety for drivers and road users.
The AIthena project and its use cases follow the AI lifecycle of design, development, deployment, operational use, monitoring, evaluation and analysis.
“The purpose of this first document of WP5 is to define the general conditions for the validation and testing activities within the AIthena project. As the report focuses on the four AIthena use cases and is published at an early stage in the project, the use cases were described first. They cover aspects of four main domains where AI is gaining importance in modern vehicles: perception, situation awareness/understanding, decision-making, and traffic management. All four use cases also relate to specific phases of the AI lifecycle. Together, the four use cases cover all of these phases. This is also described in the deliverable.
From the descriptions of the use cases, important factors can be derived like the requirements (especially the technical ones) that must be fulfilled for individual use cases. In addition, the benchmark scenarios and KPIs are defined as they are necessary for the further testing and validation process. All this information is documented in D5.1 and forms the basis for further work in WP5 as well as its further deliverables D5.2 at the end of the first project cycle and D5.3 at the end of the second project cycle.”
Bernhard Hillbrand, Virtual Vehicle (VIF) – Lead Partner of Deliverable 5.1 and leader of Work Package WP5 – DEPLOY & TEST: AI-based CCAM Deployment – integration on operation and end user validation.
To learn more about the AI and CCAM testing and methodology used in the AITHENA project, read the project deliverable here D5.1 Testing and evaluation methodology for AI-driven CCAM systems.