Provably Efficient Exploration in Reward Machines with Low Regret