Whole - ArDl Manipulation : Kinematics and Control

This thesis explores the kinematics and control issues associated with an approach to manipulation that employs all of the available surfaces of a robot to interact with the workspace. This thesis explores manipulations of this type which we term whole-arm manipulation (Wi\.LVI). A classification of WAM tasks into pushing, searching, enclosure, and exclusion is presented and examples of each are given. A technique for describing the task kinematics and compliance in terms of compliant line motions is presented. Compliance control requirements are discussed in light of these task requirements. Implementation of the kinematic solutions and control for whole-arm tasks is described for the MIT-vVAM robot. The inverse kinematic problem of placing the last link of the robot along a desired line is derived and the resulting transformations are applied to a maximum path deviation algorithm to plan manipulator motions. The endtip cartesian kin.emati{:s is derived and applied to the problem of sensing the location and Inagnitude of a cOhtact force. Experimental results are presented for an approach to contact which uses joint torque information to determine contacts. Finally, we descri be an application of the system to the whole-arm task of searching for a set of objects. Thesis Supervisor: Dr. Kenneth Salis bury Research Scientist Laboratory of Artificial Intelligence