Trustworthy AI: Landscaping Verifiable Robustness and Transparency” webinar

TUPLES presented at the Adra-e webinar  “Trustworthy AI: Landscaping Verifiable Robustness and Transparency”

Matteo Pozzi, CEO of Optit, one of our industrial partners, presented the TUPLES project during the “Trustworthy AI: Landscaping Verifiable Robustness and Transparency” webinar. This webinar is the first in a series of events promoted by Adra-e called “Birds of a Feather,” aimed at providing insights into EU Research and Innovation Actions (RIAs) projects, foster collaboration and offer a joint platform for debate and sharing results in AI, Data, and Robotics.

Matteo highlighted the key challenges of trustworthy AI in one of the five use cases of our project: the BELUGA™ Airbus Scheduling and Logistics in Aircraft Manufacturing. This use case addresses the problem of integrating production demands with the availability of Beluga flights, jigs, and trailers to optimize manufacturing logistics scheduling.

Challenges in the Use Case:

– Logistic Planning: Production demands and Beluga flights are uncertain, currently planned manually by 2-3 teams using large rack space buffers to absorb uncertainty.

– Scalability: The system must generate or update the logistics plan for 3-4 weeks of operation within 15 minutes.

– Robustness: In the event of disruptions, the system must propose a feasible plan with a high success probability in less than 15 minutes.

– Explainability: The system needs to provide different options and explain the consequences of what-if scenarios to the involved workforce.

Approaches

Supervised Learning to learn PDDL heuristics (GOOSE algorithm)

Reinforcement learning for policy learning

Psychology study on planning workers’ perspectives and needs from an AI support tool

Minimum Unsatisfiable Property Subset Explanations

The future of trustworthy AI will involve highly interdisciplinary work. The human factor is crucial in every trustworthiness framework, especially in explainability, where it plays the most significant role and cannot be addressed solely through technical approaches. Multidisciplinarity is key: the TUPLES project leverages collaboration among AI researchers from multiple universities, three industrial enterprises, ethical advisors, and the psychology department of the University of Bologna.

We are very grateful to Adra-e for promoting this meeting among projects that share our challenge of developing solutions in the field of trustworthy AI.

See the webinar here https://bit.ly/4aLs8yc