Last year in my college machine learning class, we had extra credit projects to build reinforcement learning models that were graded on their ability to play Flappy Bird. I just graduated this year and I would like to do more challenges in reinforcement learning, but Kaggle only seems to have challenges for supervised/unsupervised learning.
I am looking for something like Kaggle, but instead of competing on predictive accuracy in classification, you would be competing on net return in a simulated environment. I have seen HN discussions[1] about bot vs. bot competitions, but I haven't actually found any in public.
[1] https://news.ycombinator.com/item?id=29985837