TUPLES (TrUstworthy Planning and scheduling with Learning and ExplanationS) is a 3-year project that will contribute to a more integrated and human-centered approach to the development of P&S tools, in order to increase confidence in these systems and accelerate their adoption.
Our ambition is to obtain scalable, yet transparent, robust and safe algorithmic solutions for planning & scheduling, by designing methods that combine the power of data-driven and knowledge-based symbolic AI.
TUPLES will release a toolkit including a set of simulation environments (The Tuples Lab) for simplified versions of the case studies, an implementation of the algorithms (Scikit Environment) developed in the project, and a self-assessment tool to validate the confidence level of a planning and scheduling application.
Tuples has 3 main goals
First objective is to develop hybrid planning and scheduling methods that combine the efficiency, flexibility, and adaptability of data-driven learning approaches with the robustness, reliability, and explainability of model-based reasoning methods. This will require the ability to integrate learned models into the core of current planning and scheduling approaches that rely on constraint satisfaction, combinatorial optimization, and heuristic search algorithms.
Finally the third goal is to demonstrate these approaches on five case studies:
- Aircraft manufacturing,
- Aircraft flying assistance
- Soccer team recruitment
- Waste collection
- Energy planning
TUPLES involves four leading AI laboratories (ANITI, KU Leuven, Unibo, Saarland University) and top scientists (including 2 AAAI fellows and 2 EurAI fellows) in planning & scheduling, explanations and knowledge representation & reasoning, constrained optimisation, and hybrid AI.
It also involves Social Sciences and Humanities (SSH) researchers with backgrounds as diverse as law, philosophy, human factors and psychology, who will drive some of the research, evaluation and guidelines development.
Core to TUPLES are three small and large companies (Airbus, Optit, SciSports) developing cutting-edge AI DSS in several important industry sectors: manufacturing, aviation, energy, waste transport, sport management.
Communaute d’Université et d’etablissements Université Federale de Tolouse Mid-Pyrenees
Katholieke Universiteit Leuven
Saarland University – UdS
Universität des Saarlandes
Alma Mater Studiorum Università di Bologna
AIRBUS OPS LTD.
AIRBUS Operation Limited
WP1 ensures the ethical review of the project.
WP2 deals with the technical and administrative coordination of the project.
- In WP3, a range of new hybrid planning and scheduling methods will be developed, which enable the integration of learnt and expert knowledge about the P&S problem, and facilitate the verification and explanation of the solution in WP4 and WP5.
- WP4 will focus on the design of new efficient methods for guaranteeing a priori, and for verifying or testing a posteriori, the robustness and safety of the solution plans, policies, and schedules against both symbolic and learnt environment models.
- WP5 targets new, efficient and interactive methods for explaining the solutions produced by hybrid planners, and the legal and ethical aspects of such explanations.
In WP6 the novel methods, developed in WP3-5, will be implemented and demonstrated on our 5 use-cases, and evaluate them with our target properties (robustness, safety, transparency, scalability, human agency, environmental and societal well-being).
WP6 experience will be useful to derive practical guidelines concerning the design and development of trustworthy P&S systems, and for the development of TUPLES toolkit (diagnostic tool, TUPLES lab), design the international research competition, and integrate our hybrid P&S approaches into open-source libraries and platforms.
WP7 deals with the exploitation of the results, and organise for their wide dissemination and communication.
We adopt a use-case driven process aiming to provide structure to the research activities and ensure their practical viability. The process is schematically depicted in the figure, and will contribute both usable demonstrators for the considered use cases and simplified laboratory environments suitable for controlled experiments and public release.
TUPLES will build on its industry participants’ use cases to define how and when the AI techniques developed in the project should be applied during the various stages of implementing trustworthy P&S systems.
The use cases have been selected because of their potential to contribute to progressing the state of the art, their potential of advancing the state of the practice for robust and transparent scheduling and planning systems, and their manageable risk.
With input from Human Factors and Organizational Psychology experts, the project will propose metrics and protocols to assess and monitor the trustworthiness of the developed P&S systems, accounting for workers’ and stakeholders’ needs and opinions.