Vision-based stratified robotic manipulation

This paper addresses a three-dimensional implementation that results in several separate manipulators, each with intermittent contact with a central object of known geometry, cooperatively manipulating the object to a desired new position and orientation. Rolling control of an object using redundant contact of several independent manipulators is made difficult by the hybrid nature of this system that introduces complexity to the trajectory-planning problem, and by imperfection in the kinematic models which are needed to achieve such trajectory planning. The former is a theoretical problem which has been solved using Lie-algebra-based strategies to plan motion for the stratified systems. Imperfection in the kinematic models, on the other hand, leads to a practical implementation problem since even small errors in the equations that relate the internal pose of the robot to the position of the end-effector make precise sustained contact with an object difficult. The consequent lack of control of contact force combined with frictional unpredictability associated with rolling, causes gradual growth in the disparity between actual and calculated position and orientation of the object. A robust means for applying vision to compensate for imperfections in the holonomic kinematics of the robots as well as to update estimates of the pose of the object is outlined. Experimental results are also presented.