Object-Aware Regularization for Addressing Causal Confusion in Imitation Learning