Improved Regret Bounds for Undiscounted Continuous Reinforcement Learning

We consider the problem of undiscounted reinforcement learning in continuous state space. Regret bounds in this setting usually hold under various assumptions on the structure of the reward and transi