Ph.D., Purdue University, Dec. 1997
Co-Major Professors: Anthony A. Maciejewski and V. Balakrishnan
Robot failures are not uncommon, with consequences ranging from economic impact in industrial applications to potentially catastrophic incidents in remote and/or hazardous environments. In most instances, the failure is limited to an individual joint and joint brakes can be employed to isolate the failure. However, significant disruption to the tool trajectory and reduction in the manipulator workspace can result. One approach towards addressing this problem is to employ kinematically redundant manipulators, those with more degrees of freedom than necessary for the desired task. Kinematically redundant manipulators can achieve end-effector positions within the workspace boundaries of the robot with an infinite number of joint configurations, thus allowing the optimization of secondary criterion, such as tolerance to locked-joint failures. The fault-tolerance measure used is a worst-case quantity, given by the minimum, over all single joint failures, of the minimum singular value of the post-failure Jacobians. The computation required to compute this measure with brute-force methods is prohibitive for real-time implementation. The main contributions of this work are algorithms (based on the power method and Taylor series) that quickly compute approximations of the fault-tolerance measure and its gradient. These, in turn, result in control schemes that are implementable in real-time and are optimally fault-tolerant to single locked-joint failures. Statistical comparisons, over a wide range of manipulator designs, show that the accuracy of the best technique (the power method) is indistinguishable from that of brute-force implementations. An example demonstrating real-time fault-tolerant control, as well as optimal fault-tolerant end-effector design and workpiece placement, is presented using a commercially available seven degree of freedom manipulator.