Sensor processing for mobile robot localization, exploration and navigation

摘要

In the context of a mobile robotic agent, we describe a unified framework for the competencies of localization, navigation, exploration and map-building. In this work, we focus on localization. We describe the design, implementation, testing and evaluation of data-processing algorithms for three sensor modalities: ultrasound, stereo vision and patterned light. We develop a sensor model for each modality. In each case, distinctive features of objects in the field of view are extracted. In each case, the output of the algorithm is of a form which facilitates integration within the framework, and hence the localization of the agent in a partially-known environment. We delineate a computationally efficient method for sensor-based localization of a mobile robot based on planar features extracted using ultrasound. The method runs in time linear in the number of detected features, both for establishing correspondences between extracted and map features, and for pose estimation. We outline the extension of the localization algorithm to integrate the other sensor modalities. Finally, we describe a method which, given an input of pose measurements by the sensor-based localization algorithm, produces the minimax risk fixed-size confidence set estimate for the pose of the mobile agent. The work in this dissertation constitutes the first application to the mobile robotics domain of optimal fixed-size confidence-interval decision theory. The approach is evaluated in terms of theoretical capture probability and empirical capture frequency during actual experiments with the mobile agent. The method is compared to several other procedures including the Kalman Filter (minimum mean squared error estimate) and the identity estimator which uses the measurements itself as the estimate. The minimax approach is found to dominate all the other methods in performance.