Improving Certified Robustness In Adversarial Reinforcement Learning