---
format: "aidr-story-markdown/v1"
id: "48c1e843d4f1bfc1ff6ebcb050011c59f970d7f1e3125090266aff2e5702eba2"
canonical_url: "https://aidr.today/48c1e843?lang=en"
title: "LeWAM: End-to-End World and Action Modeling with JEPAs"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-14T00:00:00.000Z"
category: "Research"
topics: ["agent"]
source_urls: ["https://huggingface.co/papers/2610.12818"]
summary: "LeWAM is an end-to-end JEPA for control that jointly trains next latent state prediction and action flow matching with an isotropic Gaussian constraint on the latent state, improving success rates on contact-rich and long-horizon tasks while planning faster than prior JEPAs."
---

# LeWAM: End\-to\-End World and Action Modeling with JEPAs

> [Open the canonical story](<https://aidr.today/48c1e843?lang=en>)

**Published:** 2026-10-14T00:00:00.000Z
**Category:** Research
**Topics:** agent

## Summary

LeWAM is an end\-to\-end JEPA for control that jointly trains next latent state prediction and action flow matching with an isotropic Gaussian constraint on the latent state, improving success rates on contact\-rich and long\-horizon tasks while planning faster than prior JEPAs\.

## Sources

- [Story source](<https://huggingface.co/papers/2610.12818>)