Reinforcement Learning on Damaged Models
Keywords
No Thumbnail Available
Authors
Issue Date
2019-02-05
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
I test Reinforcement Learning on models that have been damaged to
see if model functionality re-emerges after training. Two di erent ap-
proaches are used to compare between damage types and relearning capa-
bilities. A comparison is made between a rehabilitation and a prosthetic
scenario. First results show that damaged models can be trained back to a
functionality near to pre-damage levels. Results show that Rehabilitation
training is faster and more complete than Prosthetic training. Prosthetic
training is unable to make the full model adjustments that make Reha-
bilitation training more successful. In both cases a considerable amount
of functionality is reestablished.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen