Skip to main content

Fundamentals of Intelligent Reliable Robot Systems

Objective

The basic aim of the FIRST Action was to integrate components of sensing, planning and control in order to increase the reliability and flexibility of robotic systems. The use of sensors is an essential component of a system that must plan and execute actions intelligently and reliably. Intelligent manipulation must be based on the relationship between non-local sensing, such as vision and hand-eye coordination, and between local sensing and grasping. Control actions need to be related to task descriptions and the shape representations that support planning. .
The integration of sensing, planning and control in robots was studied in order to further the development of more reliable robotic systems. This involved examining the fundamentals of task planning, geometric reasoning, control, and the use of different sensory modes, such as seeing, grasping and touching. Work has concentrated mainly on vision, and has been most notable for 3 principal advances. Firstly, the development of both algebraic and semidifferential invariants. One important new class of algorithms that will find wide applicability in reliable robotics systems is a technique for computing various forms of symmetry despite perspective distortion. Secondly, techniques for recognition of objects from a database that is essentially constant time. Thirdly, the construction of a graphical symbolic representation of the information in a scene that can support reasoning for assembly.

The most important class of robot motions in future manufacturing and processing applications are constraints. More versatile and higher level specification of constrained (or 'compliant') motions has been introduced. The framework has been used to generate and extend the tracking capabilities of constrained motion controllers. A system has been developed to program robot motions in complex environments automatically. Visual guidance for collision-avoidance is important and has been the subject of collaborative work. An efficient approach for planning the motion of six dof manipulators operating in complex and dynamic environments has been developed. The method is implemented within a computer aided design (CAD) system. Finally, a method of automatic grasp planning for a 2-dimensional articulated hand has been invented. A new capability has been added by work on transitional control between compliant motion and free motion.
APPROACH AND METHODS
Research proceeded on pairwise interactions between sensing, planning and control. This approach is a step beyond work restricted to one or two of these topics, and is expected to result in more reliable and intelligent systems.
PROGRESS AND RESULTS
-Sensing and planning. Work has concentrated mainly on vision, and has been most notable for four principal advances: (i) the development of both algabraic and semi-differential invariants in joint work between Leuven, Oxford, and Grenoble, leading to an international meeting on behalf of Esprit and its subsequent publication as a book by MIT Press. One important new class of algorithms that will find wide applicability in reliable robotics systems is a technique for computing various forms of symmetry d espite perspective distortion; (ii) techniques for recognition of objects from a database that is essentially constant time, that is, independent of the size of the database; and (iii) the construction of a graphical symbolic representation of the information in a scene that can support reasoning for assembly.
-Planning and control. The most important class of robot motions in future manufacturing and processing applications are compliant, in the sense that they must comply with a set of different constraints. For example, positional constraints and force cons traints are applied in grinding, polishing and machining. The Leuven group have introduced the "feature frame" approach as an extension to the "compliance frame" or "task frame" approach for more versatile and higher level specification of constrained (or "compliant") motions. The framework has been used to generate and extend the "tracking" capabilities of constrained motion controllers. In collaborations between Karlsruhe and Grenoble, a system has been developed to automatically program robot motions i n complex environments.
-Sensing and control. Visual guidance for collision-avoidance is an important step towards the sensor guided planning part of First and has been the subject of collaborative work by Karlsruhe, Oxford, Genoa and Grenoble. An efficient approach for plannin g the motion of six dof manipulators operating in complex and dynamic environments has also been developed. The method is implemented within a CAD system. Finally, a method of automatic grasp planning for a two-dimensional articulated hand has been invented. A new capability has been added to the consortium by work at Oxford on transitional control between compliant motion and free motion.
POTENTIAL
Sensing and planning systems for fine-motion strategies are likely to be useful in reliable assembly. The work on object modelling and visual feature extraction will be important in extending the current capabilities of industrial robots.

Coordinator

THE CHANCELLOR, MASTERS AND SCHOLARS OF THE UNIVERSITY OF OXFORD
Address
Parks Road
OX1 3PJ Oxford
United Kingdom

Participants (4)

Institut National Polytechnique de Grenoble
France
Address
46 Avenue Félix Viallet
38031 Grenoble
KATHOLIEKE UNIVERSITEIT LEUVEN
Belgium
Address
Celestijnenlaan 300B
3030 Heverlee
UNIVERSITÄT KARLSRUHE
Germany
Address
Kaiserstraße 12
76131 Karlsruhe
Università degli Studi di Genova
Italy
Address
Via All'opera Pia 11A
16145 Genova