Question on how Experience Replay solves the traditional instability of online Q-learning (benefits of Experience replay)

Question

My question is in regards to how Experience replay solves a traditional instability problem of RL combined with NN. Experience replay solves several instability issues, but