Box pushing with a mobile robot using visual servoing

Pushing is one of the many ways to manipulate an object and it is especially useful when the object is too big to be gripped by the robot. Previous studies analyzed the problem of pushing polygonal objects into desired poses and proposed open loop pushing algorithms. In the absence of object pose feedback, to avoid slip in contact between the robot and the object, primitives for path planning were computed using conservative estimates for coefficient of friction between the robot and the object. In this work, we experimentally measured the coefficient of friction between the robot and object to compute path planning primitives. We used A* search and RRT* algorithms for path planning. We perform controlled pushing using object pose feedback obtained from a vision system using fiduciary markers. Pushing objects with object pose feedback enables us to confidently operate close to the frictional limits of the system, as the robot can take paths with tighter turns (smaller turning radius) to push the object into the goal pose.