TOWARDS AUTONOMOUS MARS ROVER LOCALIZATION: OPERATIONS IN 2003 
MER MISSION AND NEW DEVELOPMENTS FOR FUTURE MISSIONS 
K. Di 3, *, J. Wang 3 , S. He 3 , B. Wu a , W. Chen 3 , R. Li 3 , L. H. Matthies b , A. B. Howard b 
Mapping and GIS Laboratory, CEEGS, The Ohio State University, 470 Hitchcock Hall, 2070 Neil Avenue, Columbus, 
Ohio 43210, U.S.A. {di.2, wang.813, he.l 19, wu.573, chen.1284, li.282}@ osu.edu 
b Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California 91109, U.S.A. 
{lhm, abhoward}@robotics.jpl.nasa.gov 
Commission I, ICWG I/V 
KEY WORDS: Absolute Orientation, Bundle Adjustment, Close Range Photogrammetry, Computer Vision, Matching, Navigation 
ABSTRACT: 
During the 2003 Mars Exploration Rover (MER) mission, onboard rover localization has been performed primarily by IMU, wheel- 
odometry, and sun-sensing technologies. In cases where the rover experiences slippage caused by traversing loose soil or steep 
slopes, particularly in a crater, the onboard visual odometry (VO) technique is applied. A bundle adjustment (BA) method has been 
performed on Earth to achieve a high-accuracy solution of rover positions by building and adjusting an image network containing all 
panoramas and traversing images along the entire traverse. An innovative method has been developed to automate cross-site tie- 
point selection so that BA-based rover localization can be performed autonomously onboard the rover. Recent results of MER 
mission operations and field test results are reviewed to demonstrate the effectiveness of this autonomous rover localization 
technology. 
1. INTRODUCTION 
In a planetary rover mission, localization of the rover with a 
high degree of accuracy is of fundamental importance both for 
safe rover navigation and for the achievement of science and 
engineering goals (Arvidson et al., 2004). During the 2003 
Mars Exploration Rover (MER) mission, IMU, wheel odometry, 
and sun-sensing technologies are all being used to estimate 
rover positions and attitudes within a nominal accuracy of 10 
percent. As of April 16, 2008 (Sol 1524 for Spirit; Sol 1503 for 
Opportunity), Spirit has traveled 6.67 km while Opportunity has 
traveled 11.09 km (actual distances traveled, not odometry 
measures). Onboard visual odometry (VO) is being used to 
track terrain features appearing in sequential images in order to 
correct errors caused by wheel slippage (Maimone et al., 2007). 
Due to limitations in computational speed, VO has only been 
applied to relatively short distances where the rovers have 
traveled on steep slopes or across loose soils, for example. 
In support of MER mission operations, researchers at the 
Mapping & GIS Lab of The Ohio State University (OSU) have 
been collaborating with JPL and other mission teams in 
performing bundle adjustment (BA)-based rover localization 
and topographic mapping since the landing of the two rovers in 
January 2004 (Li et al., 2005; Di et al., 2008). This BA 
technology uses tie points to link images taken at different rover 
locations, thereby forming an image network and allowing 
adjustment of the image orientation parameters to improve 
localization accuracy. Topographic maps, rover traverse maps, 
and updated rover locations have been produced and distributed 
to the science and engineering team members through a 
WebGIS for science analysis, long term planning and mission 
operations (Li et al., 2007a). 
The key to the success of BA-based rover localization is 
selection of a sufficient number of well-distributed tie points for 
linking the images along the rover traverse. Autonomous rover 
localization requires full automation of tie-point selection. From 
the beginning of MER operations, tie points linking a stereo 
pair (intra-stereo tie points) and tie points linking adjacent 
stereo pairs within one panorama taken at one rover location 
(inter-stereo tie points) were selected automatically during 
MER operations. However, cross-site tie points (ones that link 
panoramas taken at different rover locations) were selected 
manually during MER operations for the first three years. 
Recently, we developed an innovative approach to automatic 
cross-site tie-point selection so that BA-based rover localization 
can be autonomously performed onboard the rover (Li et al., 
2007b). The new approach has been verified using actual Spirit 
rover data as well as field test data acquired at Silver Lake, 
California. This new autonomous BA software has been applied 
in MER operations since August 2007. 
2. MARS ROVER LOCALIZATION DURING MER 
MISSION OPERATIONS 
At the Gusev Crater landing site, localization of the Spirit rover 
has been performed sol by sol based on incremental bundle 
adjustment using full or partial Navcam/Pancam panoramic 
images along with, occasionally, forward- and backward 
looking Navcam/Pancam middle-point survey images. The 
achievable localization accuracy has been evaluated based on a 
consistency check of the BA results. Overall, after BA, 2D 
accuracy generally ranged from less than 1 up to 1.5 pixels 
while 3D accuracy was at a centimeter to sub-meter level (Li et 
al., 2005; Di et al., 2008). Figure 1 shows the Spirit bundle- 
adjusted traverse map as of Sol 1524 in the area of Home Plate. 
Corresponding author.