Humans often push when grasping or lifting is inconvenient or infeasible, because pushing requires fewer contacts and fights against only a fraction of the object's weight. However, pushing results are hard to predict, because the physical parameters that govern the pushing motion are difficult to measure. We derived a physics-based box pushing model and implemented a feedback-based pushing pipeline using the model. Experimental results show that our pushing model has fair predictive power and our pushing pipeline can reliably push the target to the goal. We compared our physics-based method to a minimalistic baseline pushing method and showed that our method is more accurate and reliable.