AdaJEPA: An Adaptive Latent World Model

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Ying Wang, Oumayma Bounou, Yann LeCun, Mengye Ren
• Venue: arXiv: cs.LG
• Published: 2026-06-30
• Collected: 2026-07-01
• URL: http://arxiv.org/abs/2606.32026v1
• PDF: https://arxiv.org/pdf/2606.32026v1
• DOI: UNKNOWN

Plain-Language Brief

The paper proposes a new adaptive latent world model called AdaJEPA that can improve planning accuracy by continuously recalibrating the world model during test-time, particularly when dealing with distribution shift. This is achieved by using self-supervised adaptation signals and performing a closed-loop update.

What the Document Says

• Latent world models are used for planning from high-dimensional observations.
• These models are typically kept frozen at test time, but can fail when predictions become inaccurate under distribution shift.
• The paper proposes AdaJEPA, an adaptive latent world model that performs test-time adaptation within the closed loop of Model Predictive Control (MPC).
• After training, AdaJEPA plans and executes the first action chunk, uses the observed next-state transition as a self-supervised adaptation signal, and replans with the updated model.
• This closed-loop update continuously recalibrates the world model without additional expert demonstrations.
• AdaJEPA improves planning success significantly with as few as one gradient step per MPC replanning step across a range of goal-reaching tasks.

Key Concepts Mentioned

• Latent World Models
• Adaptive Latent World Model
• Test-time Adaptation
• Model Predictive Control (MPC)
• Self-supervised Learning
• Distribution Shift
• Planning Success

Claims Made by the Document

• AdaJEPA can improve planning success significantly.
• AdaJEPA can continuously recalibrate the world model without additional expert demonstrations.

Evidence Used

• The paper reports substantial improvement in planning success with AdaJEPA.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Latent world models enable planning from high-dimensional observations by predicting future states in a compact latent space. However, these models are typically kept frozen at test time: when their predictions become inaccurate, planning can fail, especially under test-time distribution shift. To address this, we propose AdaJEPA, an adaptive latent world model that performs test-time adaptation within the closed loop of model predictive control (MPC). After training, AdaJEPA plans and executes the first action chunk, uses the observed next-state transition as a self-supervised adaptation signal, and replans with the updated model. This closed-loop update continuously recalibrates the world model without additional expert demonstrations. Across a range of goal-reaching tasks, AdaJEPA substantially improves planning success with as few as one gradient step per MPC replanning step.