Imitation Learning in Continuous Action Spaces: Mitigating Compounding Error without Interaction