While studying for an RL course, I created a reference for several algorithms with a brief description of what limitations they solve. Example:
Problem: SARSA pushes q-values towards the current policy, but ideally we'd want optimal values.
Solution: Use the best action in TD-target calculation -> Q-learning
Perhaps someone else will find it helpful!
