Issue No. 02 - April (1995 vol. 10)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/64.395356
<p>Is there a robust basis for dexterous manipulation tasks? This approach relies on reusable control laws to put together manipulation strategies on-line.</p> <p>A reasonable goal in the design of robot systems is to enhance the autonomy of manipulation strategies. Although dexterous hand/arm systems provide mechanical flexibility, they present formidable control challenges that are not amenable to traditional methods such as model- and behavior-based approaches (see the sidebar). This is primarily because these systems are highly redundant, are applied to tasks composed of multiple goals, operate in uncertain and unstructured environments, and must capture a broad range of control contexts.</p> <p>To address these issues, we developed a "bottom-up" approach that composes behavior on line from a set of reusable feedback control laws called a control basis. Literally, a control basis is set of elemental controllers that collectively span a task domain under linear composition. The interaction of the control basis and a composition policy determines the control actions of the composite controller. This controller inherits predictability directly from the robustness and stability of the elemental controllers. Each elemental controller independently regulates a subset of the system's degrees of freedom, producing a control structure with distributed computational and kinematic resources.</p> <p>We constructed a control basis and composition policies for reaching, grasping, and manipulation. Demonstrations of these tasks show that the approach scales well and the underlying controllers are reusable.</p>
K. Souccar, J. A. Coelho Jr., R. A. Grupen and M. Huber, "Distributed Control Representation for Manipulation Tasks," in IEEE Intelligent Systems, vol. 10, no. , pp. 9-14, 1995.