This is the Reinforcement way for solving the problem of evacuation from a room through one door. This problem aims to capture the different paths which each pedestrian has used. From these routes, the machine can understand the place of obstacles too. My way of compelling each player to remain in the path is to put food on it and put enemies out there. The ultimate & biggest food is on the door sites indeed. There are also some other enemies too: in each step, each player's site is a place of penalty/enemy for everyone else. So each player should learn to avoid the others. The hesitation penalty is too defined to force players to take the fastest route in time and distance.
The Reinforcement ideas and base codes are from the tutorials on web-page https://pythonprogramming.net/machine-learning-tutorials/ . Thanks to them, this is a good tutorial, and I recommend it to you if you are willing to learn and understand the theory and coding for RL.