Behavior-Invariant Task Representation Learning with Transformer-based World Models for Offline Meta-Reinforcement Learning

arXiv 2026 59.4 method

TLDR

Offline meta-RL framework using Transformer-based world model and invariant task representation to handle distribution shifts and sparse rewards.

Reasoning

The paper proposes a novel integration of information-theoretic task representation with a Transformer-based stochastic world model to address context and policy distribution shifts in offline meta-RL. Strengths include clear problem formulation and extensive evaluations showing superior performance, but weaknesses include lack of real-world experiments and reliance on simulated environments.

Read-first score

Read-first score 59.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

92.1

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.92103099

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,dataset,evaluation,metric,result

Topical relevance 29%

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 386.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

generative world model

world simulator

interactive world model

video world model

Deep Analysis

Innovations

Information-theoretic task representation learning that extracts behavior-invariant latent variables
Transformer-based stochastic world model for offline meta-reinforcement learning
Conservative value penalty on imagination-based rollouts to mitigate policy shift and model exploitation

Methodology

The framework integrates information-theoretic task representation learning with a Transformer-based stochastic world model. It extracts task-defining latent variables that are invariant to the behavior policy to address context distribution shift, and applies a conservative value penalty to imagination-based rollouts to handle policy shift and model exploitation. The approach is evaluated on offline meta-reinforcement learning tasks with baselines and metrics for stability and generalization.

Key Results

The method outperforms state-of-the-art approaches, demonstrating superior stability and generalization under out-of-distribution and sparse-reward settings.

Resources

Paper