Graduation Degree: 
AA Ph.D.
Graduation Year: 


sgd AT stanford DOT edu

Education History

B.S., Physics, Carnegie Mellon University, 2005
B.S., Mechanical Engineering, Carnegie Mellon University, 2005
M.S., Aeronautics and Astronautics, Stanford University, 2008
Ph.D., Aeronautics and Astronautics, Stanford University, 2016

Current Project

Students in the Aerospace Robotics Lab have the tremendous privilege of working with the Monterey Bay Aquarium Research Institute (MBARI).  Together with MBARI, we have pushed the development of Terrain Relative Navigation (TRN) algorithms for AUVS.  It's been a great introduction to the interesting world of underwater robots.

The focus of my research is improving the robustness of TRN algorithms.  TRN is susceptible to 'false fixes' - overconfidence in erroneous estimates.  As the purpose of a filter is to produce a consistent estimate and associated untertainty, this is a big problem!

My research has had two main goals: avoiding false fixes to begin with, and recognizing when there is a false fix.  The first problem has been tackled by looking at the underlying causes of false fixes - in my case, trying to understand how a false fix can arise in 'flat' terrain.  I found that error in the map can cause TRN failure, due to the way TRN measurements are correlated.  The result of this research is a change in measurement weighting dependent on estimated terrain information. The second goal, recognizing false fixes, requires monitoring filter consistency.  At a basic level, it consists of evaluating whether the recorded measurements are consistent with the expected measurement.  Disagreements may be caused by sensor error, map error, or false fixes.  Current work is towards developing and implementing consistency checks suitable for TRN filters. 

Gradient-Based TRN

Most TRN techniques rely on correlating measurements of altitude against an stored elevation map.  Gradient based TRN extends this ability to vehicles without altitude measurements.  With gradient TRN, the fundamental measurement is the local slope.  As the robot drives along, it measures roll and pitch angles, which can be correlated against a differentiated elevation map.

Gradient TRN can enable map based localization to novel environments, e.g. the Moon, Mars, or GPS denied environments on earth.  Gradient TRN, however, has a significant SNR challenge.  Differentiating an elevation map amplifies errors present in the original DEM.  Current TRN algorithms are susceptible to false fixes in low information/low SNR environments. On of the benefits of working with rovers is I get to test the algorithms on our ATRV-Jr!  It's a great excuse to head out of lab for a nice walk - if you're interested in my research, you're welcome to join me!

Associated Projects

Research Interests

I like robots!

Last modified Tue, 2 May, 2017 at 10:30