More than a Million Ways to Be Pushed: A High-Fidelity Experimental Data Set of Planar Pushing

Pushing is a motion primitive useful to handle objects that are either too large or too heavy to grasp. It is also at the core of much of robotic manipulation, in particular when motion or reconfiguration is involved. It seems reasonable then to wish for robots to have reliable models to predict and understand how pushed objects move. However, these models comprise many parameters, such as those involving friction, that are hard to measure precisely. It is in the essence of frictional interaction and therefore of sliding that when you repeat two experiments with as close as possible to identical setups, two slightly different outcomes happen. In robotics, we often appeal to assumptions and approximations which render models that are computable, but restricted and inaccurate. Just how close are those models? How reasonable are the assumptions they are based on? To help in answering these questions, and to get a better experimental understanding of pushing, we present a comprehensive and high-fidelity data set of planar pushing experiments. The data set contains timestamped poses of a pusher and a pushed object, as well as forces at the interaction. We vary the push interaction in 6 dimensions: surface material, shape of the pushed object, contact position, pushing direction, pushing speed, and pushing acceleration. We automate the data capturing process with an industrial robot along precisely controlled position-velocity-acceleration pushing trajectories, which give us dense samples of positions and forces of uniform quality. We finish the paper by characterizing frictional properties at the interaction, as well as describing and evaluating the most common assumptions and simplifications made by models of frictional pushing in robotics.