A topological model for intelligent robot prehension

摘要

The need, in robot manipulation, for higher levels of dexterity and versatility than those provided by grippers and by special-purpose end-effectors has prompted much work on the design and control of multifingered robot hands during the last decade. Among the most prominent hand designs are the Stanford/JPL three-fingered hand, the UTAH/MIT four-fingered hand, and the USC/Belgrade five-fingered hand. Work on multifingered hand control has dealt with low-level (numeric) control, mostly based on screw theory and tools drawn from line-geometry, differential geometry, kinematics and dynamics. AI-oriented, high-level (symbolic) control schemes based on neurophysiological, behavioral or anthropomorphic approaches have also been investigated, but separately from low-level control schemes. This thesis deals with a symbolic-numeric integrated scheme for intelligent robot prehension, i.e. for intentional grasping and dextrous manipulation of a perceived object by multifingered hands. More specifically, we develop a grasp-based technique that integrates contact-based numeric methods and task-oriented symbolic methods. Our scheme is based on a novel, topological model of multifingered robot hands. This model consists of: (i) a collection of uniform representations of the space of all hand postures, described at different levels of geometric detail, and (ii) a set of continuous mappings among those spaces. These mappings are used for the determination of hand postures and hand functionality. With techniques drawn from point-set, combinatorial, algebraic topology, combinatorial geometry, and geometric modeling, we develop a set of topological reasoning algorithms for dextrous manipulation, i.e. algorithms for the derivation of numeric joint variables from symbolic task descriptions. Based on our topological model, we also design an intelligent model-based robot prehension system and an experimental software simulation tool called DEXTROUS (Development and EXecution Tool for RObots Using Simulation). To demonstrate the power of the topological concepts, we apply our topological model to two basic problems that are considered as major challenges among researchers in dextrous manipulation: hand pre-shaping and grasp planning. Finally, we discuss possible extensions of the topological approach, the model of robot prehension, the set of algorithms for symbolic-numeric integration, and experimental results to problems involving multiple coordinated robot manipulators.