Computing on Wheels: A Deep Reinforcement Learning-Based Approach