Experiments with model-simplified computed-torque manipulator controllers for free-flying robots

Abstract

A simplified computed-torque controller is presented for two common tasks performed by a two-armed freeflying robot: manipulator inertial endpoint control and cooperative-arm object manipulation. A computed-torque controller is formulated in such a manner that the differences between fixed-base and free-flying controller formulations are explicit and well defined. An analysis of these differences shows that the terms due to robot body accelerations are computationally expensive to include in the controller model, whereas those due to robot body velocities are both computationally inexpensive and vital to proper operation of the controller. For the two robot tasks under study, numerical simulations predict a wide range of robot/payload mass distributions for which robot body accelerations can be neglected in the manipulator controller model with insignificant performance degradation in both open- and closed-loop control. The error introduced is less than that typically due to uncertainties in mass distribution (1 %). The performance of two simplified computed-torque controllers are compared to their fullorder counterparts using an experimental laboratory robot and found to be indistinguishable from one another. An estimation is made of the significant computational savings possible with this simplification.