Uncertainty Quantification in the context of 6D Pose Estimation

We use the work on Deep Evidential Regression that was initially developed in the context of simple regression in R, and extend it to work on higher dimensional groups and manifolds with a Lie algebra structure. We develop a general framework for the Deep Evidential Loss and assess how well the models perform on several manifolds, identify and discuss several limitations encountered through experiments. Finally, one of the major goals of this thesis is to assess whether we could be use the Deep Evidential method in the context of 6D Object Pose Estimation, where it can be crucial to have both information on the prediction and the uncertainty on the prediction (really important for safety-critical robotic manipulation).

Level thesis:degree_level

Masters

Department dc:contributor.department

Computer Science

Grantor dc:publisher

Georgia Institute of Technology

Year dc:date.issued

2022

Author dc:creator

• Boumerdassi, Maya

Advisor dc:contributor.advisor

• Pradalier, Cédric

Committee members dc:contributor.committeemember

• Brito, Gerandy
• Yang, Diyi

• Pose Estimation
• Uncertainty Quantification

Handle dc:identifier.uri

http://hdl.handle.net/1853/66606

OAI identifier oai:identifier

oai:repository.gatech.edu:1853/66606