Date of Award
Spring 2019
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Mechanical Engineering
First Advisor
Schimmels, Joseph M.
Second Advisor
Huang, Shuguang
Third Advisor
Rice, James
Abstract
Current industrial robotic manipulators, and even state of the art robotic manipulators, are slower and less reliable than humans at executing constrained manipulation tasks, tasks where motion is constrained in some direction (e.g., opening a door, turning a crank, polishing a surface, or assembling parts). Many constrained manipulation tasks are still performed by people because robots do not have the manipulation ability to reliably interact with a stiff environment, for which even small commanded position error yields very high contact forces in the constrained directions. Contact forces can be regulated using compliance control, in which the multi-directional elastic behavior (force-displacement relationship) of the end-effector is controlled along with its position. Some state of the art manipulators can directly control the end-effector's elastic behavior using kinematic redundancy (when the robot has more than the necessary number of joints to realize a desired end-effector position) and using variable stiffness actuators (actuators that adjust the physical joint stiffness in real time). Although redundant manipulators with variable stiffness actuators are capable of tracking a time-varying elastic behavior and position of the end-effector, no prior work addresses how to control the robot actuators to do so. This work frames this passive compliance control problem as a redundant inverse kinematics path planning problem extended to include compliance. The problem is to find a joint manipulation path (a continuous sequence of joint positions and joint compliances) to realize a task manipulation path (a continuous sequence of end-effector positions and compliances). This work resolves the joint manipulation path at two levels of quality: 1) instantaneously optimal and 2) globally optimal. An instantaneously optimal path is generated by integrating the optimal joint velocity (according to an instantaneous cost function) that yields the desired task velocity. A globally optimal path is obtained by deforming an instantaneously generated path into one that minimizes a global cost function (integral of the instantaneous cost function). This work shows the existence of multiple local minima of the global cost function and provides an algorithm for finding the global minimum.