Proceedings of the 1999 IEEE International Conference on Robotics & Automation Detroit, Michigan· May 1999 The Design of Control Strategies Tolerant to Unidentified Failures in Kinematically Redundant Manipulators M. Goel, A. A. Maciejewski, and V. Balakrishnan Purdue University School of Electrical and Computer Engineering 1285 Electrical Engineering Building West Lafayette, Indiana 47907-1285 ABSTRACT The use of robots in hostile environments significantly in- creases the likelihood of failures in the robot's subsystems. Exist- ing techniques for developing failure tolerant robots rely on effec- tive failure detection and identification. Since failure identifica- tion is itself a difficult process that may not always be successful, it is important to consider the behavior of the robot prior to iden- tification of a fault, or even the possibility of failures remaining unidentified. This work proposes control strategies that improve local measures of failure tolerance for kinematically redundant robots experiencing unidentified locked-joint failures. Measures to evaluate the fault tolerance capability of the various schemes are presented and the performance of the proposed schemes are demonstrated with an example. 1. INTRODUCTION fail, one common failure mode is a "locked joint" , where the affected joint's velocity is identically zero. When such a failure remains unidentified by the robot con- troller, very large and often unpredictable deflections of the robot end effector can result. Such behavior is clearly unacceptable in cluttered environments due to the increased probability of collisions, and also in the case of teleoperated systems because of operator disori- entation caused by the unexpected motions. This work focuses on developing control strategies that improve local measures of failure tolerance for kine- matically redundant manipulators experiencing uniden- tified locked-joint failures. A general class of tasks char- acterized by sequences of point-to-point moves in task space is considered. II. MATHEMATICAL FRAMEWORK The position and/or orientation (henceforth re- ferred to as "position") of the end effector of a manipu- lator can be expressed in terms of its joint variables by the kinematic equation where J E IR mxn is the manipulator Jacobian, x is the end-effector velocity, and q is the joint velocity. If perfect servo control of the joints is assumed, then in a healthy manipulator the actual joint velocity qa equals the commanded velocity qc. However, in the event of a locked-joint failure of the i-th joint, the cor- responding element of <la is identically zero. Then, the where x E JRm is the position of the end effector, q E IR n is the vector of joint variables, and m and n the dimensions of the task space and joint space re- spectively. Manipulators that have more degrees of free- dom (DOFs) than required for a task, i.e., n > m, are said to be redundant. The end-effector velocity is ex- pressed in terms of the joint rates as Robots are being increasingly used to replace humans for applications in hazardous environments. While failures are not uncommon in industrial robots [1], the likelihood of failures is far greater when robots are applied in harsh environments [2]. Since the very nature of these environments does not allow imme- diate human intervention for repair or recovery, the abil- ity of a robot to cope with the failures becomes desir- able. Common methods of making robots failure toler- ant incorporate some form of redundancy in the design; either through duplication of components [3] or through kinematic redundancy [4,5]. Existing failure-tolerance schemes, however, rely on successful failure detection and identification; only after a failure is identified is an appropriate failure recovery strategy initiated [6,7]. Since failure identification is itself a difficult process that may not always be successful [8], it is essential to consider the possibility of delayed failure identification, or even that of failures remaining unidentified, when de- veloping any effective failure-tolerant control technique. While there are several ways in which a robot may This work was supported in part by Sandia National Labora- tories under contract no. AL-3011, the Office of Naval Research under contract no. NOOOI4-97-1-0640, and NSF under contract no. MIP-9708309. x = f(q), x= Jq, (1) (2) 0-7803-5180-0-5/99 $10.00 © 1999 IEEE 867