BiTAgent: A Task-Aware Modular Framework for Bidirectional Coupling between Multimodal Large Language Models and World Models

arXiv 25.12 2025 57.7 method, system

TLDR

BiTAgent bidirectionally couples MLLMs and world models via task-aware dynamic joint learning for embodied agents.

Reasoning

The paper introduces a novel bidirectional framework that addresses key coupling and adaptability challenges, with strong methodological components. However, the abstract is cut off, limiting full assessment of results and real-world validation.

Read-first score

Read-first score 57.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 47.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%

67.1

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

Methodology quality 25%

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

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 351.

Keyword Scores

world model

world dynamics prediction

interactive world model

model-based reinforcement learning world model

generative world model

world simulator

video world model

Deep Analysis

Innovations

Bidirectional coupling between MLLMs and WMs via forward and backward pathways
Task-Aware Dynamic Joint Learning
Task-Aware Behavior Learning
MLLM-WM Joint Optimization

Methodology

BiTAgent is a task-aware dynamic joint framework that couples MLLMs and WMs through two complementary pathways: a forward path injecting MLLM representations into the WM's latent space for semantically guided imagination, and a backward path where WM-generated feedback refines the MLLM's semantic space via dense text-conditioned rewards. The framework integrates three synergistic components: Task-Aware Dynamic Joint Learning, Task-Aware Behavior Learning, and MLLM-WM Joint Optimization. Experiments are conducted across multi-task and cross-environment settings.

Key Results

Extensive experiments across multi-task and cross-environment settings demonstrate superior stability and generalization over state-of-the-art baselines.

Resources

Paper