The second practitioner’s workshop of the AIthena project was held online on 26 February 2025 (details including agenda). The workshop is a part of a series of yearly practitioner’s workshop with an aim to facilitate an in-depth exchange between stakeholders and other interested parties on the tools and methods developed in the AIthena project and providing participants with a comprehensive understanding of their practical applications and potential impact in real-world applications in CCAM systems.
In the introductory part of the workshop, the AIthena project which focuses on building trustworthy, explainable and accountable CCAM technologies was introduced. The value proposition of the project, emphasizing its potential impact and the key challenges that may be encountered during its realization were outlined. A brief overview of the project’s approach and objectives was also provided, laying the groundwork for the following discussions and presentations.
Participants:
The online workshop was attended by seventeen participants including external attendees (representing FitDrive, CONNECT, FRODDO, UP2DATE4SDV, and BERTHA projects).
Presentations:
Two presentations were given during the workshop, each covering the main aspects of the project.
1. Building a Trustworthy AI Methodology in AIthena
The first presentation focused on a trustworthy AI methodology developed as a tool in AIthena to build trustworthy AI applications in CCAM applications. This tool can serve as a checklist to enable different users like developers, regulators, authorities, testers or end users to evaluate the ethical, transparency, accountability and privacy-preserving aspects of AI applications in CCAM. The structure of the tool and how to evaluate the AI applications with respect to four trustworthiness related topics, fairness, transparency, accountability, and privacy were presented.
2. Data Anonymisation Pipelines and GDPR Compliance
In the second presentation, the importance and motivation for anonymization of recorded data and being GDPR compliant was outlined. The presentation highlighted key challenges and the importance of data anonymization at the source without loss of important information. A comparison between different anonymization methods and the advantage of using GenAI to mitigate the suboptimal results observed in other methods like blurring was explained.
Interactive sessions:
Following the presentations, the workshop featured an interactive session where the topic for discussion was ‘Addressing Real-World Challenges with the AIthena Approach’. Menti was used to post questions related to the real-world challenges in CCAM applications and how they can be prioritized. Participants provided their responses on the main challenges that they see are important while implementing the before mentioned methodology. A poll was also conducted to prioritize the main challenges wherein the ‘Availability of Experts’ received a highest number of votes and ‘Understanding Regulatory Framework’ coming in second.
In the second part of this session, practical demonstrations of the AIthena use-cases were presented. The first use case that was presented was ‘Trustworthy Perception systems for CCAM – Reliable Pedestrian Detection in Urban Environments’, focusing mainly on the pipeline and its implementation. Key challenges like discrepancies between data or detections from different sensors were also highlighted. Additionally, the use of data and model cards to evaluate fairness and other trustworthy AI aspects was discussed.
The second use case that was presented was ‘Trustworthy and Human understandable Decision Making – Explainable and robust decision making’, which highlights the importance of decision making for planning and maneuvering in autonomous driving. How ML decision-making and human understandable definitions of traffic rules encoded in the HD maps can be combined to make the system more robust and explainable to maximize safety and comfort. The concept was also illustrated with different traffic scenarios and examples. At the end, the AIthena methodology checklist was also applied to the two use-cases presented here to evaluate the AI trustworthiness aspects namely fairness, transparency, accountability and privacy.
In conclusion, the upcoming AIthena activities were mentioned where different stakeholders can collaborate for CCAM related topics and projects.