Pro-DEUCE
A Platform for Safe Reinforcement Learning and Verification with Industrial Applications. ERC Proof of Concept Grant (2026, 18 months).
Pro-DEUCE brings the results of the ERC Starting Grant DEUCE closer to practice. The project builds an open-source tool platform that connects standard reinforcement learning environments with abstraction, model learning, formal verification and planning, and then validates the resulting policies back in the original environment. The goal is to make safe reinforcement learning with formal guarantees usable for engineers outside the research community.
The platform is developed and tested together with industrial partners from the Netherlands, Germany, the UK and Sweden: Stateful Robotics (robot mission control), Rolsch Assetmanagement (sewer and drinking-water asset management) and Ericsson (telecommunication networks), with the startup incubator Cube 5 at Ruhr University Bochum exploring the route towards a spin-off.
Pro-DEUCE is funded by the European Research Council (ERC) as a Proof of Concept Grant.