Robotic Localization and Perception in Static Terrain and Dynamic Urban Environments
This dissertation presents a complete, real-time, field-proven approach to robotic localization and perception for full-size field robots operating outdoors in static terrain and dynamic urban environments. The approach emphasizes formal probabilistic yet efficient frameworks for solving salient problems related to robotic localization and perception, including 1) estimating robot position, velocity, and attitude by fusing GNSS signals with onboard inertial and odometry sensors, 2) aiding these navigation solutions with measurements from onboard landmark sensors referencing a pre-surveyed map of environmental features, 3) estimating the locations and shapes of static terrain features around the robot, and 4) detecting and tracking the locations, shapes, and maneuvers of dynamic obstacles moving near the robot. The approach taken herein gives both theoretical and data-driven accounts of the localization and perception algorithms developed to solve these problems for Cornell University's 2005 DARPA Grand Challenge robot and 2007 DARPA Urban Challenge robot. The approach presented here is divided into four main components. The first component statistically evaluates variants of an Extended Square Root Information Filter fusing GNSS signals with onboard inertial and odometry sensors to estimate robot position, velocity, and attitude. The evaluation determines the filter's sensitivity to map-aiding, differential corrections, integrity monitoring, WAAS augmentation, carrier phases, and extensive signal blackouts. The second component presents the PosteriorPose algorithm, a particle filtering approach for augmenting robotic navigation solutions with vision-based measurements of nearby lanes and stop lines referenced against a known map. These measurements are shown to improve the quality of the navigation solution when GNSS signals are available, and they keep the navigation solution converged in extended signal blackouts. The third component presents a terrain estimation algorithm using Gaussian sum elevation densities to model terrain variations in a planar gridded elevation model. The algorithm is validated experimentally on the 2005 Cornell University DARPA Grand Challenge robot. The fourth component presents the LocalMap tracking algorithm, a real-time solution to the joint estimation problem of data assignment and dynamic obstacle tracking from a potentially moving robot. The algorithm is validated in controlled experiments with full-size vehicles, and on data collected at the 2007 DARPA Urban Challenge.
This dissertation documents the perception algorithm in Cornell University's 2005 DARPA Grand Challenge robot, and the localization and perception algorithms in Cornell University's 2007 DARPA Urban Challenge robot.
This material is based upon work supported under a National Science Foundation Graduate Research Fellowship. This work is also supported by the DARPA Urban Challenge program (contract no. HR0011-06-C-0147), with Dr. Norman Whitaker as Program Manager.
DARPA Urban Challenge; DARPA Grand Challenge; Robotic Localization; Robotic Perception; Bayesian Estimation; Sensor Fusion; GNSS; Inertial Navigation; Extended Square Root Information Filtering; Hypothesis Testing; Map-Aided Localization; Particle Filtering; Terrain Estimation; Gaussian Mixture
dissertation or thesis