Multi-Agent Egocentric World Model with Fine-Grained Embodied Interaction
ME-World generates synchronized first-person streams for multiple agents using fine-grained shared-world actions.
ME-World is a multi-agent egocentric world model that generates synchronized first-person streams for agents interacting through fine-grained actions in a shared environment. It jointly denoises multiple ego streams in one token sequence, conditions each stream on all agents' target-view poses, and uses shared environment memory. Trained and evaluated on real and synthetic multi-agent data, it improves shared-world consistency, action control, identity preservation, and video quality over prior methods.
- ME-World jointly denoises multiple first-person video streams.
- Each stream is conditioned on all agents' target-view poses.
- Shared environment memory grounds generation across agents.
- New metrics cover environment, update, and identity consistency.
From the original article
1 min read at huggingface.co
Egocentric world models predict first-person observations conditioned on an agent's actions, but most focus on a single agent. Real embodied settings often involve multiple agents that act and interact within a shared environment. Existing multi-agent world models rely on coarse actions like locomotion, camera control, or discrete commands, leaving fine-grained embodied interactions underexplored.
By Dahyun Chung, Siyoon Jin, Hyunwook Choi · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.