Robot AI model
V-JEPA 2-AC
V-JEPA 2-AC post-trains an action-conditioned latent world model using DROID data and enables image-goal planning on Franka arms. Repository code uses MIT with Apache-2.0 exceptions; independently stated terms for the action-conditioned checkpoint were not established.
Records last verified Oct 13, 2026 · Send a correction
- Organization
- Meta FAIR
- Type
- World model
- Released / introduced
- Jun 11, 2025
- Public weights
- Public weights
- Licence / scope
- Not established
- Parameters
- Not disclosed
- Base model
- V-JEPA 2 ViT-g/16
- Humanoid association
- None reported
- Inputs
- vision, proprioception, actions
- Outputs
- predicted visual representations
Release dates can mark an announcement or paper introduction. Public checkpoints, code licences and weight terms are distinct; follow the publisher’s links for the stated scope.
Documented embodiments
Source-associated hardware; this list does not establish plug-and-play compatibility or commercial robot capability.
- Franka robot arm
Real-robot planning experiments documented in the repository.
Disclosed training data
More robot AI models
Evidence
Sources
- [s1] V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning ↗
Meta FAIR · paper · published Jun 11, 2025 · accessed Oct 13, 2026
- [s2] V-JEPA 2 official README: action-conditioned checkpoint and Franka planning ↗
Meta FAIR · official · accessed Oct 13, 2026
- [s3] V-JEPA 2 paper: action-conditioned model inputs and robot planning ↗
Meta FAIR · paper · published Jun 11, 2025 · accessed Oct 13, 2026