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

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 learni...

Full description

Saved in:
Bibliographic Details
Main Authors: Siddharth Padmanabhan, Kazuki Miyazawa, Takato Horii, Takayuki Nagai
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11000102/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849734985679896576
author Siddharth Padmanabhan
Kazuki Miyazawa
Takato Horii
Takayuki Nagai
author_facet Siddharth Padmanabhan
Kazuki Miyazawa
Takato Horii
Takayuki Nagai
author_sort Siddharth Padmanabhan
collection DOAJ
description 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.
format Article
id doaj-art-d77c6c8f97284b7dbf5780603413cf5b
institution DOAJ
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-d77c6c8f97284b7dbf5780603413cf5b2025-08-20T03:07:40ZengIEEEIEEE Access2169-35362025-01-0113838578386610.1109/ACCESS.2025.356878411000102Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action DataSiddharth Padmanabhan0https://orcid.org/0009-0009-3339-3123Kazuki Miyazawa1Takato Horii2https://orcid.org/0000-0001-5374-0887Takayuki Nagai3https://orcid.org/0009-0009-1503-651XGraduate School of Engineering Science, Osaka University, Toyonaka-shi, Osaka, JapanGraduate School of Engineering Science, Osaka University, Toyonaka-shi, Osaka, JapanGraduate School of Engineering Science, Osaka University, Toyonaka-shi, Osaka, JapanGraduate School of Engineering Science, Osaka University, Toyonaka-shi, Osaka, JapanRecent 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.https://ieeexplore.ieee.org/document/11000102/GPThumanoidimitation learningmotion predictionwhole-body control
spellingShingle Siddharth Padmanabhan
Kazuki Miyazawa
Takato Horii
Takayuki Nagai
Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
IEEE Access
GPT
humanoid
imitation learning
motion prediction
whole-body control
title Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
title_full Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
title_fullStr Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
title_full_unstemmed Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
title_short Data-Efficient Approach to Humanoid Control by Fine-Tuning a Pre-Trained GPT on Action Data
title_sort data efficient approach to humanoid control by fine tuning a pre trained gpt on action data
topic GPT
humanoid
imitation learning
motion prediction
whole-body control
url https://ieeexplore.ieee.org/document/11000102/
work_keys_str_mv AT siddharthpadmanabhan dataefficientapproachtohumanoidcontrolbyfinetuningapretrainedgptonactiondata
AT kazukimiyazawa dataefficientapproachtohumanoidcontrolbyfinetuningapretrainedgptonactiondata
AT takatohorii dataefficientapproachtohumanoidcontrolbyfinetuningapretrainedgptonactiondata
AT takayukinagai dataefficientapproachtohumanoidcontrolbyfinetuningapretrainedgptonactiondata