QT-TDM:利用变压器动力学模型和自回归 Q 学习进行规划

IF 4.6 2区 计算机科学 Q2 ROBOTICS
Mostafa Kotb;Cornelius Weber;Muhammad Burhan Hafez;Stefan Wermter
{"title":"QT-TDM:利用变压器动力学模型和自回归 Q 学习进行规划","authors":"Mostafa Kotb;Cornelius Weber;Muhammad Burhan Hafez;Stefan Wermter","doi":"10.1109/LRA.2024.3504341","DOIUrl":null,"url":null,"abstract":"Inspired by the success of the Transformer architecture in natural language processing and computer vision, we investigate the use of Transformers in Reinforcement Learning (RL), specifically in modeling the environment's dynamics using Transformer Dynamics Models (TDMs). We evaluate the capabilities of TDMs for continuous control in real-time planning scenarios with Model Predictive Control (MPC). While Transformers excel in long-horizon prediction, their tokenization mechanism and autoregressive nature lead to costly planning over long horizons, especially as the environment's dimensionality increases. To alleviate this issue, we use a TDM for short-term planning, and learn an autoregressive discrete Q-function using a separate Q-Transformer (QT) model to estimate a long-term return beyond the short-horizon planning. Our proposed method, QT-TDM, integrates the robust predictive capabilities of Transformers as dynamics models with the efficacy of a model-free Q-Transformer to mitigate the computational burden associated with real-time planning. Experiments in diverse state-based continuous control tasks show that QT-TDM is superior in performance and sample efficiency compared to existing Transformer-based RL models while achieving fast and computationally efficient inference.","PeriodicalId":13241,"journal":{"name":"IEEE Robotics and Automation Letters","volume":"10 1","pages":"112-119"},"PeriodicalIF":4.6000,"publicationDate":"2024-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"QT-TDM: Planning With Transformer Dynamics Model and Autoregressive Q-Learning\",\"authors\":\"Mostafa Kotb;Cornelius Weber;Muhammad Burhan Hafez;Stefan Wermter\",\"doi\":\"10.1109/LRA.2024.3504341\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Inspired by the success of the Transformer architecture in natural language processing and computer vision, we investigate the use of Transformers in Reinforcement Learning (RL), specifically in modeling the environment's dynamics using Transformer Dynamics Models (TDMs). We evaluate the capabilities of TDMs for continuous control in real-time planning scenarios with Model Predictive Control (MPC). While Transformers excel in long-horizon prediction, their tokenization mechanism and autoregressive nature lead to costly planning over long horizons, especially as the environment's dimensionality increases. To alleviate this issue, we use a TDM for short-term planning, and learn an autoregressive discrete Q-function using a separate Q-Transformer (QT) model to estimate a long-term return beyond the short-horizon planning. Our proposed method, QT-TDM, integrates the robust predictive capabilities of Transformers as dynamics models with the efficacy of a model-free Q-Transformer to mitigate the computational burden associated with real-time planning. Experiments in diverse state-based continuous control tasks show that QT-TDM is superior in performance and sample efficiency compared to existing Transformer-based RL models while achieving fast and computationally efficient inference.\",\"PeriodicalId\":13241,\"journal\":{\"name\":\"IEEE Robotics and Automation Letters\",\"volume\":\"10 1\",\"pages\":\"112-119\"},\"PeriodicalIF\":4.6000,\"publicationDate\":\"2024-11-21\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"IEEE Robotics and Automation Letters\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://ieeexplore.ieee.org/document/10759753/\",\"RegionNum\":2,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"ROBOTICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Robotics and Automation Letters","FirstCategoryId":"94","ListUrlMain":"https://ieeexplore.ieee.org/document/10759753/","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ROBOTICS","Score":null,"Total":0}
引用次数: 0

摘要

受变形金刚架构在自然语言处理和计算机视觉领域的成功经验启发,我们研究了变形金刚在强化学习(RL)中的应用,特别是使用变形金刚动力学模型(TDM)对环境动态进行建模。我们通过模型预测控制(MPC)评估了 TDM 在实时规划场景中的连续控制能力。虽然变压器在长视距预测方面表现出色,但其标记化机制和自回归性质导致长视距规划成本高昂,尤其是当环境维度增加时。为缓解这一问题,我们使用 TDM 进行短期规划,并使用单独的 Q 变换器(QT)模型学习自回归离散 Q 函数,以估算短视距规划之外的长期回报。我们提出的 QT-TDM 方法将变压器作为动力学模型的稳健预测能力与无模型 Q-变压器的功效相结合,减轻了与实时规划相关的计算负担。在各种基于状态的连续控制任务中进行的实验表明,QT-TDM 在性能和采样效率方面都优于现有的基于变压器的 RL 模型,同时还能实现快速、计算高效的推理。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
QT-TDM: Planning With Transformer Dynamics Model and Autoregressive Q-Learning
Inspired by the success of the Transformer architecture in natural language processing and computer vision, we investigate the use of Transformers in Reinforcement Learning (RL), specifically in modeling the environment's dynamics using Transformer Dynamics Models (TDMs). We evaluate the capabilities of TDMs for continuous control in real-time planning scenarios with Model Predictive Control (MPC). While Transformers excel in long-horizon prediction, their tokenization mechanism and autoregressive nature lead to costly planning over long horizons, especially as the environment's dimensionality increases. To alleviate this issue, we use a TDM for short-term planning, and learn an autoregressive discrete Q-function using a separate Q-Transformer (QT) model to estimate a long-term return beyond the short-horizon planning. Our proposed method, QT-TDM, integrates the robust predictive capabilities of Transformers as dynamics models with the efficacy of a model-free Q-Transformer to mitigate the computational burden associated with real-time planning. Experiments in diverse state-based continuous control tasks show that QT-TDM is superior in performance and sample efficiency compared to existing Transformer-based RL models while achieving fast and computationally efficient inference.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
IEEE Robotics and Automation Letters
IEEE Robotics and Automation Letters Computer Science-Computer Science Applications
CiteScore
9.60
自引率
15.40%
发文量
1428
期刊介绍: The scope of this journal is to publish peer-reviewed articles that provide a timely and concise account of innovative research ideas and application results, reporting significant theoretical findings and application case studies in areas of robotics and automation.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信