Our new preprint is out! A Rollout-Based Approach to Real-Time Topology and Dispatch Control in the AI4RealNet L2RPN Benchmark is joint work with my PhD students David Krame Kadurha and Robbert van der Burg, and my colleague Sandjai Bhulai. It grew out of our group’s contribution as team RAD to the AI4REALNET L2RPN 2026 Challenge.

The challenge asks teams to keep a simulated electricity grid running through outages and unfamiliar operating conditions, while keeping costs down and making good use of renewable energy. Our controller combines optimization with short simulations to decide how to reconfigure the grid and adjust generation and storage. It made it through 189 of 208 final test scenarios. (We finished fourth, just one place short of the podium!) The code is available on GitHub.