Maximum Information Measure Policies in Reinforcement Learning with Deep Energy-Based Model