logo
科言猫学术研究的AI总结
language
手机端手机端
  • Phone Feature微信小程序
简体中文简体中文
notice消息通知
博客博客
delete
arrow
返回

TREND: Task-Oriented World Models for Visual Robotic Manipulation

delete2026-04-27
delete0
PRE
AI
Y
Yuxiang Zheng
鲁
鲁涛 (Tao Lu)
蔡
蔡莹皓 (Yinghao Cai)
DOI：10.1109/LRA.2026.3688085delete
delete原文链接
delete原文求助
delete分享
delete收藏
摘要

摘要

En 中文
基于模型强化学习（MBRL）的关键组件世界模型，通过建模环境实现样本高效学习。然而，对于具有连续动作空间的挑战性视觉机器人操作任务，标准图像重建和均匀采样方案效果欠佳，甚至可能阻碍学习。为解决此问题，我们提出TREND，一种面向任务的世界模型，强调机器人任务关键特征，并利用运动信息样本加速环境建模，从而提高样本效率和策略收敛性。我们的方法包含两个组件：(1) 帧间差异捕捉（IFDC）：该机制替代传统图像重建，进行全局和局部差异捕捉以强调机器人任务关键区域，使世界模型重建任务相关特征并促进任务完成。(2) 运动密度引导经验回放（MDER）：该方案采用末端执行器位置核密度估计替代均匀采样，优先选择运动信息轨迹样本，使世界模型快速学习机器人环境动力学。我们的实验表明，在十个模拟和五个真实世界视觉机器人操作任务中，TREND在样本效率和成功率方面显著优于六种现有方法。
Keyword:
Reinforcement learning
visual learning
deep learning methods

期刊

I
IEEE Robotics and Automation Letters
IF:
5.3
论文数:
1.9K
被引数:
3.9W

机构

C
chinese academy of sciences
学者数:
56.7W
论文数: 45.0W
被引数: 704
引用论文

引用论文

Efficient Deep Reinforcement Learning With Imitative Expert Priors for Autonomous Driving具有模拟专家先验的高效深度强化学习，用于自动驾驶
errIEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
IF8.9
err2023-10-01
err64
errOAAI
errHuang, Zhiyu; Wu, Jingda; Lv, Chen
err分享
err收藏
Fine-Tuning Large Vision-Language Models as Decision-Making Agents via Reinforcement Learning
errAdvances in Neural Information Processing Systems 37
IF0
err2024-01-01
err0
PREAI
errZhai,Yuexiang; Bai,Hao; Lin,Zipeng; Pan,Jiayi; Tong,Shengbang; Zhou,Yifei; Suhr,Alane; Xie,Saining; Lecun,Yann; Ma,Yi; Levine,Sergey
err分享
err收藏
MaskGWM: A Generalizable Driving World Model with Video Mask Reconstruction
err2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
IF0
err2025-06-10
err0
PREAI
errNi,Jingcheng; Guo,Yuxin; Liu,Yichen; Chen,Rui; Lu,Lewei; Wu,Zehuan
err分享
err收藏
Selective Experience Replay for Lifelong Learning
errProceedings of the AAAI Conference on Artificial Intelligence
IF0
err2018-04-29
err0
errOAAI
errDavid Isele; Akansel Cosgun
err分享
err收藏
DreamingV2: Reinforcement Learning with Discrete World Models without Reconstruction
err2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
IF0
err2022-10-23
err0
errOAAI
errMasashi Okada; Tadahiro Taniguchi
err分享
err收藏
Subjective MSE Measures
errIEEE Transactions on Systems, Man, and Cybernetics
IF0
err1986-01-01
err0
PREAI
errHans Marmolin
err分享
err收藏
Model-based Reinforcement Learning: A Survey基于模型的强化学习: 综述
errFOUNDATIONS AND TRENDS IN MACHINE LEARNING
IF25.4
err2023-01-01
err161
errOAAI
errMoerland, Thomas M.; Broekens, Joost; Plaat, Aske; Jonker, Catholijn M.
err分享
err收藏
Human-level control through deep reinforcement learning通过深度强化学习实现人类层面的控制
errNature
IF0
err2015-02-25
err0
PREAI
errVolodymyr Mnih; Koray Kavukcuoglu; David Silver; Andrei A. Rusu; Joel Veness; Marc G. Bellemare; Alex Graves; Martin Riedmiller; Andreas K. Fidjeland; Georg Ostrovski; Stig Petersen; Charles Beattie; Amir Sadik; Ioannis Antonoglou; Helen King; Dharshan Kumaran; Daan Wierstra; Shane Legg; Demis Hassabis
err分享
err收藏
Advances and challenges in learning from experience replay
errARTIFICIAL INTELLIGENCE REVIEW
IF13.9
err2024-12-20
err0
errOAAI
errNeves, Daniel Eugenio; Ishitani, Lucila; do Patrocinio Junior, Zenilton Kleber Goncalves
err分享
err收藏
Deep Reinforcement Learning for Autonomous Driving: A Survey用于自动驾驶的深度强化学习: 一项调查
errIEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
IF8.4
err2022-06-01
err965
errOAAI
errKiran, B. Ravi; Sobh, Ibrahim; Talpaert, Victor; Mannion, Patrick; Al Sallab, Ahmad A.; Yogamani, Senthil; Perez, Patrick
err分享
err收藏
学者 查看更多内容