Back to search

Virginia Tech

Deep Lunar Rock Assessment with Vision and Vibration

Abstract

dc:description.abstract

Autonomous lunar site preparation requires assessing individual rocks before excavation or relocation. Strong illumination contrast, occlusion, and burial make this difficult: observations from different viewpoints must be assigned to the same rock before its geometry can be estimated, while surface observations alone provide limited information about embedment. This dissertation develops visual methods for recovering rock identity and geometry, complemented by a separately evaluated vibration-based method for estimating burial depth. First, an RGB-plus-inverted-depth segmentation model identifies individual rocks, and calibrated projection converts their masks into metric three-dimensional surface observations. Second, ShadowCorr determines which observations belong to the same rock by projecting volumetric shadows behind visible surfaces and learning their shared spatial support. This approach does not require appearance similarity or direct surface overlap. Separate synthetic evaluations achieved average correspondence purity of 99.1% with complete masks and 98.9% with controlled mask erosion. Third, the associated observations support geometric completion using a covariance-ellipsoid prior, followed by mesh construction and volume and centroid estimation. In the four-view synthetic evaluation, matched, valid reconstructions achieved a median volume error of 16.5% and a median centroid error equal to 4.0% of the true longest rock dimension. Compared with visible-point convex-hull and alpha-shape closures, completion improved centroid accuracy while producing higher volume error. Fourth, impulse excitation and scanning laser Doppler vibrometry measure the response of partially buried specimens. A neural-network model uses vibration features and exposed geometry to estimate burial depth without measured rock mass or soil parameters as inputs. Separate laboratory experiments with 18 concrete specimens in sand achieved a root mean squared depth error of 0.94 cm under nested leave-one-rock-out evaluation, with each specimen excluded from model development when tested. Together, these studies establish methods for recovering individual-rock geometry from incomplete visual observations and obtaining complementary burial-depth information from vibration. The results support future robotic assessment before rock removal; validation with natural rocks and lunar conditions, automated measurement, and integration of the sensing methods remain future work.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy
Level thesis:degree_level
doctoral
Discipline thesis:degree_discipline
Mechanical Engineering
Department dc:contributor.department
Mechanical Engineering
Grantor dc:publisher
Virginia Tech
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Ruan, Yiyan
Chair dc:contributor.committeechair
  • Komendera, Erik
Committee members dc:contributor.committeemember
  • West, Robert L.
  • Cooper, John R.
  • Southward, Steve C.
  • Jones, Creed Farris

Subjects

dc:subject × 17

Rights

dc:rights
Statement dc:rights
  • In Copyright
Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Other
vt_gsexam:47680

Chain of custody

source
Harvested from
Virginia Tech
Base URL
vtechworks.lib.vt.edu/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Ruan, Yiyan. Deep Lunar Rock Assessment with Vision and Vibration. doctoral thesis, Virginia Tech, 2026. https://hdl.handle.net/10919/143868