Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data

Padmanabhan, Siddharth; Miyazawa, Kazuki; Horii, Takato; Nagai, Takayuki

doi:https://doi.org/10.1109/ACCESS.2025.3568784

Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data

Padmanabhan, Siddharth; Miyazawa, Kazuki; Horii, Takato et al.

IEEE Access, 2025, 13, 83857-83866

アクセス数:81件（2026-02-15 22:21 集計）

固定URL: https://hdl.handle.net/11094/103258

閲覧可能ファイル

ファイル	フォーマット	利用条件	サイズ	閲覧回数	利用開始日	説明	information
IEEEAccess_13_83857	pdf	なし	2.40 MB	35	2025.12.02

Creative Commons : 表示

論文情報

ファイル出力 EndNote Basic出力 Mendeley出力

タイトル	Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
著者	著者名 Padmanabhan, Siddharth
	NRID 1000010976187 1000010976187 ORCID 0000-0002-0276-6048 0000-0002-0276-6048 researchmap permalink kazukimiyazawa kazukimiyazawa Scopus Author ID 57219246368 57219246368 著者名 Miyazawa, Kazuki
	NRID 1000060809333 1000060809333 researchmap permalink t-horii t-horii Scopus Author ID 55842337200 55842337200 著者名 Horii, Takato
	NRID 1000040303010 1000040303010 著者名 Nagai, Takayuki
キーワード等	GPT
	humanoid
	imitation learning
	motion prediction
	whole-body control
内容	Padmanabhan S., Miyazawa K., Horii T., et al. Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data. IEEE Access 13, 83857 (2025); https://doi.org/10.1109/ACCESS.2025.3568784.
抄録	Recent advances in imitation learning have enabled robots to learn multiple tasks from large-scale datasets. However, developing a model for multi-tasking humanoid control faces significant challenges. Human kinematic data is available in open-source datasets for humanoid motion learning, but learning policies from this data requires simulation due to the lack of actions. While dynamic data can accelerate learning via supervision, datasets typically lack substantial amounts of such action labels, that are also difficult to be directly used for training due to the unique structure of each human/humanoid systems. In this study, we pre-trained a Generative Pre-trained Transformer (GPT) based model on expert policy-rollout observations only (without actions) from a humanoid motion dataset. Upon fine-tuning on a smaller dataset with both observations and action labels, we demonstrate that our GPT-based model can predict actions to achieve human-like movements faster in training than training a GPT on the entire dataset from scratch directly. Furthermore, performance evaluation based on motion generation across various behaviors showed that our approach achieves efficient learning comparable to baselines.
出版者	Institute of Electrical and Electronics Engineers Inc.
収録物名	IEEE Access
巻(号)	13
ページ	83857-83866
刊行年月	2025-05-12
言語	英語
ハンドルURL	https://hdl.handle.net/11094/103258
eISSN	21693536
関連情報 (isIdenticalTo)	DOI (出版社版) https://doi.org/10.1109/ACCESS.2025.3568784
アクセス権	open access
権利情報	This article is licensed under a Creative Commons Attribution 4.0 International License.
出版タイプ	Version of Record
カテゴリ	学術雑誌論文 Journal Article

資源タイプ	学術雑誌論文
ローカル資源タイプ	学術雑誌論文
DCMI資源タイプ	text
DCTERMS.bibliographicCitation	IEEE Access. 2025, 13, p. 83857-83866
DC.title	Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
DC.creator	Padmanabhan, Siddharth
	Miyazawa, Kazuki
	Horii, Takato
	Nagai, Takayuki
DC.publisher	Institute of Electrical and Electronics Engineers Inc.
DC.language' scheme='DCTERMS.RFC1766	英語
DC.type' scheme='DCTERMS.DCMIType	text
DCTERMS.issued' scheme='DCTERMS.W3CDTF	2025-05-12
DC.identifier	https://doi.org/10.1109/ACCESS.2025.3568784
DC.subject	GPT
	humanoid
	imitation learning
	motion prediction
	whole-body control
DC.description	Padmanabhan S., Miyazawa K., Horii T., et al. Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data. IEEE Access 13, 83857 (2025); https://doi.org/10.1109/ACCESS.2025.3568784.
DCTERMS.abstract	Recent advances in imitation learning have enabled robots to learn multiple tasks from large-scale datasets. However, developing a model for multi-tasking humanoid control faces significant challenges. Human kinematic data is available in open-source datasets for humanoid motion learning, but learning policies from this data requires simulation due to the lack of actions. While dynamic data can accelerate learning via supervision, datasets typically lack substantial amounts of such action labels, that are also difficult to be directly used for training due to the unique structure of each human/humanoid systems. In this study, we pre-trained a Generative Pre-trained Transformer (GPT) based model on expert policy-rollout observations only (without actions) from a humanoid motion dataset. Upon fine-tuning on a smaller dataset with both observations and action labels, we demonstrate that our GPT-based model can predict actions to achieve human-like movements faster in training than training a GPT on the entire dataset from scratch directly. Furthermore, performance evaluation based on motion generation across various behaviors showed that our approach achieves efficient learning comparable to baselines.
DC.rights	This article is licensed under a Creative Commons Attribution 4.0 International License.
DC.format	103258
citation_title	Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
citation_author	Padmanabhan, Siddharth
	Miyazawa, Kazuki
	Horii, Takato
	Nagai, Takayuki
citation_publisher	Institute of Electrical and Electronics Engineers Inc.
citation_language	英語
citation_date	2025-05-12
citation_journal_title	IEEE Access
citation_volume	13
citation_firstpage	83857
citation_lastpage	83866
citation_doi	https://doi.org/10.1109/ACCESS.2025.3568784
citation_public_url	https://hdl.handle.net/11094/103258

Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data

閲覧可能ファイル

論文情報

論文詳細を表示