Development and Validation of Reinforcement Learning-Enhanced Controllers for Multi-Physical Processes (MPRL2REAL)

From the work within the Flanders AI Research Program (FAIR) and the FWO-SBO project “CTRLxAI”, new insights are obtained on how to leverage reinforcement learning (RL) to improve control methods for industrial processes in a safe, robust and resource-efficient way.

However, we noticed there is still a large gap between proof of concepts validated in the lab and actual implementation in industry, especially for multi-physical processes. This project aims to bridge this gap by building a simulation environment to validate the RL enhanced controllers. This intermediate step is required to increase the confidence level and maturity level of the new RL-enhanced control techniques and assess their suitability before real-life implementation. The project will therefore lower the risk for companies and facilitate joint follow-up trajectories.

Downloads

• Report on guidelines and best practices (VUB)

• Report on guidelines and best practices (UGent)

Contacts

Leander Schietgat (VUB) - Leander.schietgat@vub.be

Researchers

Guillaume Crevecoeur (UGent) - Guillaume.crevecoeur@ugent.be

Denis Steckelmacher (VUB AI Lab)

Pieter Nguyen (Ugent MIRO)

Jeroen De Maeyer (Ugent MIRO)

Tom Lefebvre (Ugent MIRO)

Ann Nowé (VUB AI Lab)

Research Groups

VUB AI Lab

Ugent MIRO