AI Open Competitions
Within the scope of the project AI4REALNET, the organisation of three AI open innovation competitions using the digital environments Grid2Op, BlueSky, and Flatland marks a key moment to engage with the global AI developer community.
The three competitions are planned to take place in 2026, regarding the three domains of the project: Power Grid, Railway, and Air Traffic Management.
Collectively, these competitions aim to attract international participation from researchers and domain experts, produce benchmark datasets and comparative insights, expand the contributor community for these digital ecosystems, and surface innovative approaches suitable for integration into future developments.
If you don't want to miss any news, we invite you to fill in this form and stay up to date whenever new information is available.
Railway
Real-World Baselines Challenge: Dynamic Train Rescheduling under Stochastic Perturbations
4 May - 8 June
Organisers: Flatland Association | DB | SBB
Contact: Manuel Meyer (competition@flatland-association.org)
The Railway Competition addresses the Vehicle Rescheduling Problem (VRSP) in the grid world of Flatland. Navigate trains through tight schedules and breakdowns, departure delays and infrastructure failures. Participate in this challenge to help create a state-of-the-art baseline for the future of train management systems and become part of the Flatland community.
Power grid
[L2RPN 2026] The AI4REALNET Sim2Real Challenge
22 June - 16 October
Organisers: RTE | TenneT
Contact: Bruno Lemetayer (lemetayer@rte-france.com)
The Power Grid Competition will evaluate the ability of the contenders to adapt to real world conditions and develop new approaches on the Sim2Real problem. It will be hosted on https://www.codabench.org.
To get all the information and updates on the competition, go to the dedicated channel on grid2op Discord:
Discord | #l2rpn_ai4realnet_2026 | Grid2Op
Air traffic management
Real-World Baselines Challenge: Multi-Agent Air Traffic Control with Reinforcement Learning
1 July - 15 November
Organisers: TU Delft
Contact: Joost Ellerbroek (j.ellerbroek@tudelft.nl)
This challenge explores multi-agent reinforcement learning for air traffic control within the BlueSky-Gym framework. Within this setting, the developed reinforcement learning method will have to function in procedurally generated airspaces, where multiple aircraft must navigate toward their waypoints while avoiding conflicts and restricted areas.