Exploration in Reward Machines with Low Regret