The RLeap Project

Representation Learning for Acting and Planning

The RLeap project aims at bringing together two of the main current research threads in Artificial Intelligence:
  • Data-based learners capable of inferring behavior and functions from experience and data.
  • Model-based solvers capable of tackling well-defined but intractable models.
Our research is focused on the integration of learners and solvers in the context of acting, addressing the problem of learning first-order symbolic representations from raw perceptions alone without using any specific prior symbolic knowledge. Read more about the project.

Who are we?

The RLeap project started in 2020 funded by an Advanced ERC grant, headed by Hector Geffner, who in 2023 moved from Universitat Pompeu Fabra to RWTH Aachen. The research has also been partially funded by the Wallenberg Artificial Intelligence, Autonomous Systems and Software Program (WASP) and the Humboldt Foundation. Read more about our team.