Dissertation/Thesis Abstract

SLAM-based nonlinear optimal control approach to robot navigation with limited resources
by Joo, Sungmoon, Ph.D., Stanford University, 2010, 160; 3405492
Abstract (Summary)

In many practical robot applications, maps of the environment are often unavailable or outdated. Furthermore, signals from navigational aids such as the global positioning system (GPS) are frequently occluded or might not even exist, making direct measurement of the robot's position or orientation difficult. A technique called simultaneous localization and mapping (SLAM) offers an alternative means of navigation in these situations.

Often range and bearing sensors are available in SLAM applications. Recently, however, the use of bearings-only sensors, such as monocular cameras, has been actively studied for the application to small robots because fusing bearing data with measurements from additional onboard sensors provides a navigational solution that satisfies the limited payload and/or power capacity of small robot platforms.

In bearings-only SLAM, the quality of estimates is dependent on the path followed. Stochastic control theory offers a framework for calculating optimal trajectories for SLAM, but these approaches are computationally expensive and often yield suboptimal results.

This study presents a new method for computing a trajectory for bearings-only SLAM-based navigation. This method is computationally efficient and provides near-optimal results as well. We approached the problem by extending the linear quadratic dual (LQD) control method.

In the course of the extension a new nonlinear optimal control synthesis method, called the two-step method, was developed to improve optimality and to secure stability of the solution. Also a nonlinear bearings-only SLAM estimator that uses the square root unscented Kalman filter (SR-UKF) technique was designed, and then modified for computational efficiency.

Simulations were performed to validate the proposed approach. Compared to a search-based navigational method, the new method yields near-optimal results using much fewer computations. In addition, the two-step method was tested with various examples including the bearings-only SLAM-based navigation. Depending on the example chosen, simulation results showed that the two-step method improves its cost value by up to 50%.

Indexing (document details)
Advisor: Rock, Stephen M.
School: Stanford University
School Location: United States -- California
Source: DAI-B 71/04, Dissertation Abstracts International
Subjects: Robotics
Keywords: Dual control, Kalman filters, Limited resources, Robot navigation, Simultaneous localization and mapping
Publication Number: 3405492
ISBN: 978-1-109-71287-2
Copyright © 2020 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy