To facilitate automated bin picking when parts cannot be grasped, pushing actions have the potential to separate objects and move them away from bin walls and corners. In the context of the Dexterity Network (Dex-Net) robot grasping framework, we present two novel push policies based on targeting free space and diffusing clusters, and compare them to three earlier policies using four metrics. We evaluate these in simulation using Bullet Physics on a dataset of over 1,000 synthetic pushing scenarios. Pushing outcomes are evaluated by comparing the quality of the best available grasp action before and after each push using analytic grasp metrics. Experiments conducted on scenarios in which Dex-Net could not successfully grasp objects suggest that pushing can increase the probability of executing a successful grasp by more than 15%. Furthermore, in cases where grasp quality can be improved, the new policies outperform a quasi-random baseline by nearly 2 times. In physical experiments on an ABB YuMi, the highest performing push policy increases grasp quality by 24%.