LeWAM: A JEPA World Action Model with Diffusion-Steering-Based MPC
arXiv cs.AI · 9 Oct 2026, 03:46 AEST
LeWAM is a new bidirectional transformer designed for action and policy prediction in world action models. It offers improved alignment, acting, and planning capabilities compared to traditional models.
Key facts
- World action models (WAMs) predict actions and future observations, typically from a reconstruction-based representation that carries noisy, redundant information which can complicate downstream predictions.
- We introduce LeWAM, a bidirectional transformer for forward, backward, inverse dynamics and policy prediction, on a decoder-free JEPA latent trained end-to-end through all four modes.
- Closed-loop evaluations of LeWAM match a regular flow-matching policy trained on the same encoder at matched size, while also providing a world model.
- Sampling raw actions when planning with WAMs lets MPC exploit dynamics-model inaccuracies; planning in the noise space of the policy head instead improves the closed-loop performance of these WAMs.
Read aloud
LeWAM is a new bidirectional transformer designed for action and policy prediction in world action models. It offers improved alignment, acting, and planning capabilities compared to traditional models. World action models WAMs predict actions and future observations, typically from a reconstruction-based representation that carries noisy, redundant information which can complicate downstream predictions. We introduce LeWAM, a bidirectional transformer for forward, backward, inverse dynamics and policy prediction, on a decoder-free JEPA latent trained end-to-end through all four modes. Closed-loop evaluations of LeWAM match a regular flow-matching policy trained on the same encoder at matched size, while also providing a world model. Sampling raw actions when planning with WAMs lets MPC exploit dynamics-model inaccuracies; planning in the noise space of the policy head instead improves the closed-loop performance of these WAMs.
Sources in this cluster
- LeWAM: A JEPA World Action Model with Diffusion-Steering-Based MPC
arXiv cs.AI · 9 Oct 2026, 03:46 AEST