Development for Space Robotics with Emphasis on Fault-Tolerance, page 3

Decision-Making and Control - A redundant robot is an extremely complex system with essentially limitless options for performing most tasks. The extra resources demand active utilization during normal operation. This refers to the redundancy management mode of operation where the control inputs are selected based on the optimization of selected performance criteria. Our position is that no single criteria is sufficient for decision-making and control, but rather that a suite of weighted and ranked criteria must form the basis for any intelligent decision making process. Towards this goal we have conceptualized over 100 different performance criteria and mathematically formulated 30 of these. The section on criteria development describes some of them. After formulating and prioritizing the performance criteria for a given system, there still remains the problem of incorporating them into a decision making system that will maximize performance while simultaneously satisfying operational constraints. Fault-tolerance places additional demands on the decision maker because the robot may suddenly lose one or more resources, thus requiring a change in control emphasis from one of redundancy management to that of failure management. The failure management further breaks into chronological stages. First is fault detection and isolation (FDI). The fault detection routines must continuously monitor the system and upon detection of a fault, must isolate and identify the source of the fault. At this point, the reconfigurable control system responds to the change in the robot's resources and automatically restructures the control algorithm so that the control inputs are reconfigured while at the same time maintaining task performance. Condition based maintenance on the robot will restore it to full-capability.

Serial Robot with 21 Degrees of Freedom - As an example of redundancy resolution in a fault-tolerant system, consider the inverse kinematics problem for the massively-redundant serial robot with 21 degrees of freedom shown. Though this is clearly a conceptual robot, it represents a system with

Serial Robot with 21 Degrees of Freedom - As an example of redundancy resolution in a fault-tolerant system, consider the inverse kinematics problem for the massively-redundant serial robot with 21 degrees of freedom shown. Though this is clearly a conceptual robot, it represents a system with a tremendous degree of redundancy. We have developed a unique redundancy resolution technique based on the method of sequential filters developed by Eschenbach and Tesar. This method of redundancy resolution explicitly identifies a set a feasible options for the robot's motion in the next instance using a series of joint-level perturbations. Perturbing the joint displacements a potential solution set. The sequential filters then evaluate and rank the options based on the performance criteria and operational constraints. The logical sequence first applies the least computationally demanding constraints, and then evaluates the remaining options based on the higher level performance criteria. The application of equality constraints on the end-effector’s placement, followed by travel, speed, and acceleration limits at the joint-level will typically eliminate the vast majority of the options. A fault in one of the joints simply removes the options associated with perturbations of the faulty joint from the feasible set. Using this technique, we can resolve the redundancy of this extremely complex system at hundreds of cycles per second on common personal computers.

Performance Criteria - The ability of a decision making system to control a redundant robot system depends on the quality of the information it is provided to make those decisions. Performance criteria are mathematically rigorous metrics which are derived from the kinematic and dynamic robot models Each performance criterion quantifies a characteristic inherent to the operation of the robot and thereby provides vital information concerning its current state. The decision making system may make use of these criteria to determine the quality of various self-motion configurations and select a solution that best achieves the specified goals. Performance criteria may divide into two categories based on the information they require. Task dependent criteria rely on information concerning the current state of the robot’s task; such as velocity or force specifications. Conversely, task independent measures are defined only from the physical parameters inherent to the robot; geometry, inertia, and compliance for example. Next Page ->