Hi there. I’m part of the competition organising team, and helped develop the RLlib baselines. Their current status is that they’re usable, and I believe the branch containing them will also have everything set up to submit models trained with rllib for evaluation. They haven’t been published as we’re less able to support them than the torchbeast baseline, but they’re definitely usable, and we’d be very interested to see if trying out any of the other algorithms improves results over IMPALA. Let me know if you do use those baselines, and if you find any bugs we can try and fix them. Good luck in the competition!