Automatic synthesis of compliant motions for robot tasks with uncertainty

Abstract

This thesis presents a new approach to the problems of control synthesis and motion planning for compliant motions of robot manipulators. The scheme is based on decoupled linear impedance models of the robot and its environment, in which the errors due to control inaccuracy, task frame estimation, as well as environment uncertainty are accounted for. The technique consists of characterizing the position, velocity and force responses of the robot/environment system, and expressing the task goals as inequality constraints on these expressions. The impedance parameters and input trajectories of the robot controller are found by fitting a parametric uncertainty box inside the region of the parameter space that satisfies all task constraints. The box is tested for validity by comparing the response values at each of its vertices to the desired goal values, and also by verifying that it does not cross any of the frontiers that separate the eigenvalue-space in box-convex regions. A software package called Commotion has been developed which implements this methodology and searches for a set of robot impedance parameters that can perform a given task robustly. Four different task examples are presented in which the validity and usefulness of the proposed approach are demonstrated. Three of these examples have been implemented experimentally, and show the practicality of the method.