PhD Thesis Proposal: Austin Lines

Wednesday, December 4, 2019, 9:00–11:00am

Rm 132, MacLean ESC

“Advances in Autonomous Ground Traverses of the Earth's Polar Regions”

Abstract

Ground based measurements of the Arctic and Antarctica are critical in validating data from satellite and airborne sensors. The expense, logistics, and danger of collecting this data via manned traverses means that the scope and frequency of these traverses is limited. While the concept of replacing manned traverses with automated rovers has been shown to be an alternative, several improvements need to be made to render a robot as a robust and viable replacement of humans in this role. The work done and the proposed work focus on three crucial advances in automated ground traverses. The first improvement is in processing large datasets of ground penetrating radar (GPR) data collected on long distance traverses. Typically, GPR data has been analyzed using manual layer picking techniques which is time consuming and subject to human bias and error. I developed a hybrid layer picking method that allows for automated tracing of layers of interest, with human input only required for correcting areas of low signal to noise ratio data. The second area of improvement is in modifying the tractive element design of the robot to allow for maximum mobility in the majority of snowpack conditions using existing terramechanics models. This work resulted in new wheels for a robot prototype that have been shown to enhance mobility on snow. The final focus of improvement is in incipient immobilization detection using real time data collected by a suite of proprioceptive sensors. The proposed approach of detection is developing canonical nonlinear dynamical models by enhancing existing models with real observations of nonlinear effects from vehicle-terrain interactions. By matching the hypothesized models to the current robot's state and propagating the model forward by a few time steps, this method could predict a loss of traction before it occurs. The planned work involves collecting data with an instrumented robot on relevant terrains while concurrently augmenting existing models until immobilization can be reliably predicted using this method.

Thesis Committee

For more information, contact Daryl Laware at daryl.a.laware@dartmouth.edu.