Do we REALLY need an external world model? ๐ค
Do we REALLY need an external world model? ๐ค
Standard approaches often rely on heavy external simulators.
We agree with the view: The Agent itself is the World Model.
๐ How to align agentic world models via experience learning?
We are excited to introduce our new work: โAligning Agentic World Models via Knowledgeable Experience Learningโ(WorldMind)๐
๐งThe Problem: LLMs possess vast semantic knowledge but lack physical grounding. โ Ask for a plan: It sounds logical. โ Execute it: It fails physically (e.g., trying to slice without a knife). ๐ตโ๐ซ
The agent knows what to do, but not how physical laws constrain it.
๐กThe Solution: WorldMind
We bridge the gap between high-level reasoning and physical reality through:
๐ Agentic World Model: Instead of external engines, we activate the agentโs internal ability to simulate environmental dynamics to guide planning.
๐น Online Experience Learning: Eliminates the need for costly fine-tuning or retraining. ๐น Alignment via World Knowledge: Autonomously builds a World Knowledge Repository (WKR) to ground the agent.
This unifies: โข Process Experience: Learning from step-level prediction failures โ โข Goal Experience: Distilling shortcuts from successful trajectories โ
๐ Key Features:
โ Training-Free: Aligns agents via online experience learning.
โจ Superior Performance: improvements on EB-ALFRED & EB-Habitat.
๐ Project Page: https://zjunlp.github.io/WorldMind/ ๐ Paper: https://huggingface.co/papers/2601.13247
Our current method is limited by todayโs foundation models and cannot yet support reliable long-horizon planning.
Looking ahead, as model capacity and memory modules continue to improve, we believe agents will gradually internalize world models and achieve robust long-term embodied decision-making.
EmbodiedAI MultimodalAgent ExperienceLearning Alignment WorldModels LLM Robotics AgenticAI NLP WorldMind
๐ ์๋ณธ ๋งํฌ
๋ฏธ๋์ด
