QHyer model enhances offline goal-conditioned RL with adaptive history compression

Researchers have developed QHyer, a novel approach for offline goal-conditioned reinforcement learning that addresses challenges posed by partially observable and history-dependent datasets. QHyer utilizes a Q-estimator to guide policy stitching and a hybrid Attention-Mamba backbone for adaptive history compression. Experiments show QHyer achieves state-of-the-art performance on both non-Markovian and Markovian datasets. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a new method for goal-conditioned reinforcement learning that improves performance on complex datasets.

RANK_REASON This is a research paper detailing a new method for reinforcement learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• QHyer
• Decision Transformer
• LSDT
• arXiv

• paper
• other