Hi Aydar, it’s true that action-state input will allow a continuous actions selection, but you can’t perform maxarg over such continuous set of actions, as its infinite.

Overall, Q-Learning is only meant to deal with a fixed set of possible actions.