Improving Deep Reinforcement Learning Methodology for Autonomous Defense and Escort of Military High-value Assets