Researchers from CMU and Peking Introduces 'DiffTOP' that Uses Differentiable Trajectory Optimization to Generate the Policy Actions for Deep Reinforcement Learning and Imitation Learning