Georgia Institute of Technology
Space Robot Apprenticeship: Characterizing Human Rover Planners to Enable Robotic Decision Support in Path Planning on Mars
Abstract
dc:description.abstractRovers have been operated on Mars over the last 20 years to determine whether past life has ever existed on Mars, characterize Martian climate and geology, and prepare for future human exploration. Each and every day, a dedicated team of engineers called Rover Planners at the National Aeronautics and Space Administration’s Jet Propulsion Laboratory manually examine the Mars terrain around these rovers to determine what drive path the rover should take. While JPL engineers have sought to develop algorithms that autonomize the way rovers drive, these algorithms do not match the quality of human-made path plans with the appropriate prioritization to safety, efficiency, risks, and mission constraints. Within this thesis, I worked directly with an expert Rover Planner at NASA’s Jet Propulsion Laboratory to extract a dataset for previous Curiosity path plans over a select 22 Martian days, or Sols, that were generated by Rover Planners. To develop an algorithm to learn, understand, and imitate RP path planning decision-making, I conducted a series of interviews with an expert rover driver at JPL and developed metrics with careful consideration to a Rover Planner’s path planning values and priorities. With these metrics, I developed an initial planning recommendation system to provide decision support for path planning by generating a suggested rover path plan following a framework that captures operator’s objectives. An experienced Rover Planner at the Jet Propulsion Laboratory judged my system's path to be of comparable quality to a Rover Planner's path and validated that it would have been a path able to be approved to drive on Mars. Following these results, I performed various unsupervised and supervised machine learning techniques on my dataset to explore research questions regarding strategies that individual or groups of Rover Planners may employ while path planning. Lastly, to be able to conduct future tests on a physical system, I also contributed to building the JPL Open-Source Rover. This thesis was an exploratory activity that yielded important insights and informs future work on how to more closely capture appropriate prioritizations to safety, efficiency, risks, and mission constraints to be able to integrate in the Rover Planning workflow. Overall, this thesis lays out a foundational framework to enable robots to encode mental models and strategies for how humans want robots to autonomously, safely, and efficiently path plan.
Degree
thesis:*- Level thesis:degree_level
- Masters
- Department dc:contributor.department
- Electrical and Computer Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2021
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Duong, Van Ai
- Advisor dc:contributor.advisor
-
- Gombolay, Matthew
- Committee members dc:contributor.committeemember
-
- Klein, Benjamin
- Chernova, Sonia
- Lightsey, E. Glenn
Subjects
dc:subject × 4Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- http://hdl.handle.net/1853/66476