An approach to reducing parameter uncertainty for robotic surface assembly tasks.

摘要

In contrast to hard automation, which relies on the precise knowledge of all parameters and special-purpose machinery, the goal of flexible assembly is to overcome the inherent uncertainty in the location of the parts. The main result of this dissertation is that, for rigid, non-deformable objects, more accurate estimates of parameters, which describe the position and orientation in Cartesian space, can be obtained through active part interaction. If the objects have large polyhedral or convex features, the parameter estimation problem can be recast in terms of fitting the collected empirical data to a suitable geometrical model. The planning and execution steps are treated as conceptually separate from the estimation. Grasping is outside of the scope of this work, and it is assumed throughout that one of the objects is held by the gripper. Additionally, an algorithm for automatic conversion of a compliant path from the Cartesian to the joint space of a general-purpose, 7 degrees of freedom robotic arm is described. This allows for the assembly strategies to be planned in terms of objects' topological features in the task frame. A `back-drivable' Barrett WAM robotic arm without a force sensor was used in all experiments. Consequently, the primary focus is on planning, control, and assembly without force sensing. Instead, approximate compliant motion is achieved by relying on torque limits and impedance. The underlying concepts, however, are much more general and could be extended to incorporate force feedback.