Dynamic robot multisensor fusion/integration system for flexible automation.

摘要

The objective of this dissertation is to address the major issues of robot multi-sensor fusion/integration for the application of flexible automation.; We have described a hierarchical four-phase approach which termed as "Far Away," "Near To," "Touching" and "Manipulation" for acquiring robot sensory information. The information obtained in each phase were then represented into distinct template. Object template can be obtained by combining each distinct sensing templates. We have modeled the sensors using probability distribution function. We also developed a scheme for Confidence Distance measure and sensor error detection before actual fusion process occurred. To estimate the optimum fused sensor data, a general paradigm using Fisher model, static approach, or Moving Quadratic curve fitting, dynamic approach in local sensor estimation, preprocessing, and Bayesian model in optimum estimation has been explored.; Sets of simulated sensor data have been generated to test the proposed multi-sensor fusion paradigm. These simulated sensor data are generated by different types of random number generator to simulated various types of sensor characteristics. Furthermore, a sophisticated Unimate PUMA 560 robot equipped with various external sensors also has been used to test the success of the entire system. The external sensors include, overhead vision, eye-in-hand vision, tactile array, position, force/torque sensing devices.