Using Logical Specifications of Objectives in Multi-Objective Reinforcement Learning