Stereo-Vision Localization and Mapping on Embedded Systems for Planetary Micro-Rover Navigation Using GPU Computed Characteristic Features
- Paper number
IAC-12,D3,2,2,x13100
- Author
Mr. Cameron Frazier, Faculty of Engineering, University of Ottawa, Canada
- Coauthor
Mr. Robert Hewitt, Faculty of Engineering, Queen's University, Canada
- Coauthor
Prof. Alex Ellery, Carleton University, Canada
- Coauthor
Dr. M. John D. Hayes, Faculty of Engineering, Carleton University, Canada
- Year
2012
- Abstract
The computational power of current planetary rovers is constrained by many factors including restrictive power sources, limited mass allowances and device radiation susceptibility. % These computational limits hinder the abilities of small, efficient and low-cost micro-rovers to effect autonomous navigation or other complex actions -- path planning, scientific investigation and sensor fusion capabilities are all affected. % The use of simultaneous localization and mapping (SLAM) for navigation is a common component of autonomous rover control. It is expensive in terms of computational and power requirements due to the volume of data processing and the demands of the sensing devices. % To permit a reduction in both computational and power requirements placed upon the micro-rover, the image characteristic features are processed via the graphical processing unit (GPU) of a low-powered system-on-a-chip (SoC) computing platform. % This paper will outline the methods and implementation of this system using a commercial micro-rover platform; a commercial low-power SoC with an embedded GPU; image processing for characteristic features using said GPU; a SLAM implementation based upon FastSLAM2. % A working and testable SLAM implementation is created for an existing rover chassis using commercial, low-power SoCs; the computational system power demands are reduced from the baseline system; the computational performance is increased from the baseline system; a functioning, autonomous micro-rover platform is demonstrated.
- Abstract document
- Manuscript document
(absent)