I am relatively new in this field, but I couldn't find anything similar to this problem.
The problem: An agent can move from state s1 to state s2 in many ways (in one step). For example if states represent locations, assume that an agent can move from location represented by s1 to the one of s2 in one step, by taking one of the actions a1 or a2. This means that multiple actions taken in some state lead to the same state.
Is there anything similar in the literature?
Yes, this situation it pretty standard and can be managed by any Reinforcement Learning algorithm. Markov Decission Processes (which is the mathematical framework commonly used to model the environment in RL) do not assume that there is a unique action that can lead from one state s1 to a another state s2.
So any literature about RL is also covering the case you describe.
For example, this MDP from Wikipedia article for Markov decision process shows an case where you can move from state s1 to state s2 in two ways and one step:
