If you are interested in this kind of problems, I encourage you to check out the Flatland competition: <a href="https://www.aicrowd.com/challenges/flatland-challenge" rel="nofollow">https://www.aicrowd.com/challenges/flatland-challenge</a><p>The idea is to optimize the scheduling of Swiss trains, using either operations research (OR) or reinforcement learning (RL).<p>Currently, they only use OR, however as the scale of the network and the number of trains keeps growing, the hypothesis is that RL may soon provide better approaches - they won't necessarily be globally optimal, but they will be good enough, and faster to re-calculate if something goes wrong.<p>The competition is currently in-between rounds but it will re-open soon. The idea is to keep it running "forever" as a long-term OR vs RL benchmark.<p>Disclaimer: I am currently affiliated with AIcrowd.