The AIthena consortium would like to share with you the progress of work done in the first year of AIthena’s implementation. You find below a summary of work in the AIthena project until November 2023.
WP1 METHODOLOGY: AI Human-centric approach
Exciting developments are happening in Work Package 1. AIthena is crafting a methodology to assess and support the development of human-centric AI-based Cooperative, Connected, and Automated Mobility (CCAM) functions. We have just wrapped up the first draft of a tool designed to assist CCAM developers, testers, and anyone diving into CCAM functions. What’s the tool about? It provides checklists and guidance to help ensure CCAM functions are fair, transparent, accountable, explainable, and rock-solid on data privacy. And there’s more. We asked for and received input from across Europe to a user survey, conducted in ten different languages. We did that to delve deeper into the hearts and minds of people, and try to understand their needs, concerns, and expectations about self-driving vehicles. This invaluable knowledge is now feeding our AIthena use cases, making sure they shine as prime examples of human-focused, AI-based CCAM.
WP2 DATA: Life cycle management and generation
During the first 13 months of Work Package 2, the consortium dedicated its efforts to the development of a first version of a driving dataset. This encompassed the meticulous design, preparation, and generation of an initial dataset utilizing an array of sensors, including cameras, lidars, and radars. Notably, this dataset incorporates both real-world data, captured through diverse sensor types, and synthetically generated data derived from virtual scenarios. To enhance the accessibility and transparency of the dataset, the consortium introduced a Data Card template. This template serves as a structured mechanism for exhaustively documenting both publicly available and newly created datasets. Emphasizing data provenance and governance principles, these Data Cards facilitate a detailed understanding of the datasets, enabling Machine Learning processes to effectively interpret and utilize the data. Moreover, recognizing the importance of privacy in handling sensitive information, the consortium developed various privacy anonymization tools. These tools are instrumental in obscuring private areas within images, such as faces and license plates. This ensures a heightened level of privacy protection while maintaining the integrity and utility of the dataset for research and development purposes.
WP3 XAI: Explainable Continuous Model Development
During this time, we have been reviewing the stages of an ML (Machine Learning) model development pipeline and exploring the different tools that can support developments in each of these stages. Also, we defined complete and standardized ways to communicate useful information about the ML models from the developers to the stakeholders. In this effort, we have designed an ML framework tools diagram to visualize the different tools involved in creating the model. Also, a model card in which details about the creation of the model, its intended use, limitations, and other useful information are reported. These documents contribute to the transparency of the ML model lifecycle. Additionally, we have been focusing on the development of first versions of AI algorithms for perception using fusion, making use of amongst other hybrid AI models and real-synthetic data for edge-case training, as well as improving trustworthy decision making, with a first use case on robo-taxi as a concrete example which will be further integrated in Work Package 5.
WP4 TOOLS: and testing facilities
Over the past 13 months, our work package has been dedicated to establishing the needs, requirements, and architecture of a novel toolchain. This toolchain is designed to improve data handling within Cooperative, Connected and Automated Mobility (CCAM), integrating insights from various project phases to streamline Artificial Intelligence model testing and validation. Our approach involved a blend of physical and virtual testing methods and the creation of a cloud-based infrastructure. This infrastructure is key for managing extensive datasets, vital for scalable and efficient AI-driven systems. These efforts have laid the groundwork for more robust and reliable data processing in our field.
WP5 DEPLOY & TEST: AI-based CCAM Deployment – integration on operation and end user validation
Work Package 5 has just started and is about testing and deploying the AIthena use cases (more about UCs). These activities are divided into two cycles (in the middle and at the end of the project). The first tasks will be the development of a validation approach including the definition of benchmark scenarios, key performance indicators and requirements. The AI lifecycle of the use case activities will also be considered, as well as the evaluation of AI limitations and conflicts in the use cases.
WP6 IMPACT: Exploitation, Dissemination and Standardisation
During the first year of the project, in WP6 we have developed the AIthena branding guidelines, recommendations for consortium partners when it comes to communication activities, and guidelines for dissemination of the results. We have created tools for visibility of the project by producing the project printed materials, setting up the AIthena website and social media channels: LinkedIn, and X (Twitter) to exchange with stakeholder community. AIthena partners have brought the project objectives, goals and first achievements to several international discussion forums (conferences, symposia, events), presenting scientific papers and posters, and discussing the AIthena relevant developments within the CCAM community. Additionally, to further discuss and develop the IMPACT activities AIthena consortium partners met in person for the occasions of the kick-off meeting in Brussels, and two consortium meetings: in San Sebastian (hosted by Vicomtech), and in Eindhoven (hosted by Eindhoven University of Technology). Finally, we organised the 1st AIthena workshop which was focused on the results of the survey developed in WP1, and the key concepts of Data and Model Cards developed in WP3.
Stay tuned for more updates in AIthena project!