Relative Position Sensing by Fusing Monocular Vision
and Inertial Rate Sensors
Andreas HusterJuly 2003
A dissertation submitted to the department of Electrical Engineering
and the Committee on Graduate Studies of Stanford University
in partial fulfillment of the requirements
for the degree of Doctor of Philosophy.
Abstract
Sensing the relative position between a robot and objects in its environment is a core requirement for many robot tasks and is an area of active research. This dissertation describes the development of a new, robust, relative-position sensing strategy suitable for unstructured and unprepared environments. Underwater manipulation, the use of underwater vehicles to perform object manipulation tasks in the ocean, is the particular application that motivated this research. Although many relative position sensing systems have already been developed, achieving the level of robustness that is required for operation in the underwater environment is very challenging.The new sensing strategy is based on fusing bearing measurements from computer vision and inertial rate sensor measurements. These measurements are fused to compute the relative position between a moving observer and a stationary object. Inertial rate sensors are composed of accelerometers and rate gyros.
The requirements on the vision system have been chosen to be as simple as possible: tracking a single feature on the object of interest with a single camera. This is equivalent to a bearing measurement. Simplifying the vision system has the potential to create a more robust sensing system. The relative position between a moving observer and a stationary object is observable if these bearing measurements, acquired at different observer positions, are combined with the inertial rate sensor measurements, which describe the motion of the observer.
The main contribution of this research is the development of a new, recursive estimation algorithm which enables the sensing strategy by providing a solution to the inherent sensor fusion problem. Fusing measurements from a single bearing sensor with inertial rate sensor measurements is a nonlinear estimation problem that is difficult to solve with standard recursive estimation techniques, like the Extended Kalman Filter (EKF). A new, successful estimator design---based on the Kalman Filtering approach but adapted to the unique requirements of this sensing strategy---was developed. The new design avoids the linearization of the nonlinear system equations. This has been accomplished by developing a special system representation with a linear sensor model and by incorporating the Unscented Transform to propagate the nonlinear state dynamics.
The dissertation describes how the sensing strategy can be implemented to determine the relative position between a moving observer and a stationary object. A demonstration task is developed that illustrates how a real-time implementation of this sensing strategy can be incorporated into the closed-loop control of an autonomous robot to perform an object manipulation task. The performance of the sensing strategy is evaluated with this hardware experiment and extensive computer simulations. Centimeter-level position sensing for a typical underwater vehicle scenario has been achieved.
PDF Download
Entire dissertation: dissertation.pdfIndividual Sections:
- Frontmatter
- Chapter 1. Introduction
- Chapter 2. System Equations
- Chapter 3. Recursive Nonlinear Estimation
- Chapter 4. Estimator Design
- Chapter 5. Experimental System
- Chapter 6. Results
- Chapter 7. Conclusions
- Bibliography
If you have any comments, send me an e-mail.