A quasi-static model-based control methodology for articulated mechanical systems

Hazardous environments encountered in nuclear clean-up tasks mandate the use of complex robotic systems in many situations. The operation of these systems is now performed primarily under teleoperation. This is, at best, five times slower than equivalent direct human contact operations. One way to increase remote work efficiency is to use automation for specific tasks.; This research considers a quasi-static macroscopic modeling methodology that could be combined with sensor-guided manipulation schemes to achieve the needed operational accuracies for remote work task automation. Application of this methodology begins with an off-line analysis phase in which the system is identified in terms of the ideal D-H parameters and its structural elements. The manipulator is modeled with fundamental components (i.e. beam elements, hydraulic elements, etc) and then analyzed to determine load dependent functions that predict deflections at each joint and the end of each link. Next, forces applied at the end-effector and gravity loads are projected into local link coordinates using the undeflected pose of the manipulator. These local loads are then used to calculate deflections which are expressed as 4 by 4 homogeneous transformations and inserted into the original manipulator transformations to predict end-effector position and orientation (and error/deflection vector).; Real-time compensation strategies have been developed so as to lessen concerns with structural deformation during use. The compensation strategies presented here show that the modeling methods can be used to increase the end-effector accuracy by calculating the deflections and command adjustments iteratively in real-time. The iterations show rapid convergence of the adjusted command positions to reach the desired end-effector location. The compensation methods discussed are easily altered to fit systems of any complexity, only requiring changes in the number of variables and the number of equations to solve. Most importantly, however, is that the modeling methodology, in conjunction with the compensation methods, can be used to correct for a significant fraction of the errors associated with manipulator flexibility effects. Implementation in a real-time system only involves changes in path planning, not in low-level control.; The modeling methods and deflection predictions were verified using a sub-system of the Oak Ridge National Laboratory's Dual Arm Work Platform. The experimental method used simple, non-contact measurement devices that are minimally intrusive to. the manipulator's workspace. The results show good correlation between model and experimental results for some configurations. (Abstract shortened by UMI.)...