Model-Based Imitation Learning with Accelerated Convergence