CEM-GD: Cross-Entropy Method with Gradient Descent Planner for Model-Based Reinforcement Learning