REXASI-PRO is divided into 8 Work Packages.
Except WP1 dedicated to the general coordination of the project the other WPs are described below:
Requirements for the development of Reliable & Explainable AI (ReXASI) framework based on swarm AI with the human in the loop
The Work Package aims at gathering the requirements for the development of a trustworthy solution by design. It will involve all partners in defining a set of system requirements that, following the 7 EU guidelines lead to the development of a trustworthy AI.
Design of trustable solutions for social robotic navigation
The Work Package aims to develop algorithms that increase trust in robots assisting users navigating among people, resulting in robots that:
- move legibly, efficiently and safely;
- cooperate with people to unravel difficult, ambiguous navigation situations;
- are robust being trained on real data and using multiple sensors optimally; and
- offer a resilient dialogue-based interface.
Formal Verification and Validation for the development of robust, safe, and secure ReXASI framework
During the Work Package our techniques will target the following key components: motion control, routing, social navigation, sensing, secure communication and speech-based HMI. We will focus on certifying collision freedom on the trajectories generated using some neural network controller and enforcing safety at different levels.
Explainable AI methods to increase the reliability of AI systems
The Work Package has for objective Combining Explainability and Reliability to increase Humans and System safety. Explainability of AI modules provides humans with understanding and trust in model outcomes. Reliability of AI modules identifies methods to design data-driven safety envelopes with guarantees, such as statistical zero error or conformal predictors.
Decision Science and Topology-based methods for Greener AI
During the Work Package we will develop a suite of solutions to tame power consumptions and computational complexity of AI systems. First, the orchestrator will optimize the fleet so that it minimizes the overall energy consumption. Second, the orchestrator will employ AI algorithms that generate accurate solutions while using reduced computation requirements.
Integration and Demonstration of ReXASI-PRO for people with permanent and temporary reduced mobility
The Work Package will implement three demonstrators focused on showing the capabilities of the technologies developed thanks to the trustworthy by design approach. A detailed description is provided in the Use Case section.
Exploitation, dissemination, and communication: paving the way for REXASIPRO’s acceptance
The Work Package will aim to identify market prospects for REXASI-PRO, thus taking care to disseminate the results of the scientific research and to assess the applicability of the solution in different social contexts.