I am working to use DQN and Policy Gradient reinforcement learning models to solve classic maze escaping problems.
So far, I have been able to train a model, which, after around 100 episodes, quickly explored ONE optimal solution to escape mazes.
However, it is easy to see that for many maze designs, the optimal solutions could be multiple, and I would like to take one step further to collect all optimal and distinguishable solutions.
However, I tried some searches online and so far, the only material I can find is this Learning Diverse Skills. But this seems an obstacle to me. I somewhat believe this seems a classic (?) and an easier problem that should be addressed in the textbook?
Could someone shed light on this matter?