Biomimetic Sensing for Robotic Manipulation

In manipulation tasks, humans have the advantage over machines due to an unparalleled ability to process information from various inputs, including touch. A set of four robot end-effectors was equipped with force sensors to provide haptic feedback to aid in performing the manipulation tasks of rotating a sphere and a cube. The motion planning algorithm used to compute the robots' joint angles is called steering-using-piecewise-constant-inputs and is applicable to underactuated, nonlinear, nonholonomic, driftless systems. Nonholonomic constraints arise during contact, requiring the fingers to only roll relative to the object. However, the algorithm gives rise to new vector fields called Lie brackets that allow the fingers to be reconfigured without releasing the object, effectively increasing the workspace of the manipulation system. Experiments were conducted with fixed-point manipulation to produce a baseline for comparing reconfigurable manipulation experiments. Both open loop and closed loop, reconfigurable manipulation experiments were conducted on a spherical object. For the open loop cases, the entire trajectory was computed offline and executed on the robots as position commands to each of the joints. For the closed loop cases, the force sensors provided information to a fuzzy controller which periodically checked the grasp's quality. The force sensors also updated the algorithm with the finger's contact locations. In both forms, the reconfigurable manipulation experiments increased the system's workspace over that for fixed-point manipulation. Furthermore, the closed loop system proved to be more robust than the open loop system. This was shown by its improved repeatability and its improved performance when rotating about an arbitrary axis. An approach to switching between faces on a nonsmooth polygonal object while Lie bracketing was verified. To do this requires discernment of the edge, and the sensors used were found to be adequate for this task. In addition, it was shown that end-effectors with a compliant surface could be used to grasp the cube on its edges as an aid in manipulation. While the experiments were successful, the complexity of performing Lie bracket motions coupled with the small movements they give rise to was not conducive to manipulations requiring large object displacements. However, the method would be applicable for fine-scale, dextrous manipulations.