Saved in:
Bibliographic Details
Main Authors: Guo, Lu, Shan, Yixiang, Zhu, Zhengbang, Liang, Qifan, Song, Lichang, Long, Ting, Zhang, Weinan, Chang, Yi
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2507.15356
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911066930282496
author Guo, Lu
Shan, Yixiang
Zhu, Zhengbang
Liang, Qifan
Song, Lichang
Long, Ting
Zhang, Weinan
Chang, Yi
author_facet Guo, Lu
Shan, Yixiang
Zhu, Zhengbang
Liang, Qifan
Song, Lichang
Long, Ting
Zhang, Weinan
Chang, Yi
contents Offline reinforcement learning (RL) enables agents to learn policies from fixed datasets, avoiding costly or unsafe environment interactions. However, its effectiveness is often limited by dataset sparsity and the lack of transition overlap between suboptimal and expert trajectories, which makes long-horizon planning particularly challenging. Prior solutions based on synthetic data augmentation or trajectory stitching often fail to generalize to novel states and rely on heuristic stitching points. To address these challenges, we propose Retrieval High-quAlity Demonstrations (RAD) for decision-making, which combines non-parametric retrieval with diffusion-based generative modeling. RAD dynamically retrieves high-return states from the offline dataset as target states based on state similarity and return estimation, and plans toward them using a condition-guided diffusion model. Such retrieval-guided generation enables flexible trajectory stitching and improves generalization when encountered with underrepresented or out-of-distribution states. Extensive experiments confirm that RAD achieves competitive or superior performance compared to baselines across diverse benchmarks, validating its effectiveness.
format Preprint
id arxiv_https___arxiv_org_abs_2507_15356
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RAD: Retrieval High-quality Demonstrations to Enhance Decision-making
Guo, Lu
Shan, Yixiang
Zhu, Zhengbang
Liang, Qifan
Song, Lichang
Long, Ting
Zhang, Weinan
Chang, Yi
Artificial Intelligence
Offline reinforcement learning (RL) enables agents to learn policies from fixed datasets, avoiding costly or unsafe environment interactions. However, its effectiveness is often limited by dataset sparsity and the lack of transition overlap between suboptimal and expert trajectories, which makes long-horizon planning particularly challenging. Prior solutions based on synthetic data augmentation or trajectory stitching often fail to generalize to novel states and rely on heuristic stitching points. To address these challenges, we propose Retrieval High-quAlity Demonstrations (RAD) for decision-making, which combines non-parametric retrieval with diffusion-based generative modeling. RAD dynamically retrieves high-return states from the offline dataset as target states based on state similarity and return estimation, and plans toward them using a condition-guided diffusion model. Such retrieval-guided generation enables flexible trajectory stitching and improves generalization when encountered with underrepresented or out-of-distribution states. Extensive experiments confirm that RAD achieves competitive or superior performance compared to baselines across diverse benchmarks, validating its effectiveness.
title RAD: Retrieval High-quality Demonstrations to Enhance Decision-making
topic Artificial Intelligence
url https://arxiv.org/abs/2507.15356