Factored Latent Action World Models
Classification
View four quadrants- Major category
- Related resources
- Quadrant
- Not applicable
- Architecture
- Not applicable
- Prediction paradigm
- Not applicable
- Subcategories
- Latent-action & representation methods
- Source review status
- Not assigned
Category review. An adjacent latent-action representation study that factorizes inverse/forward dynamics and uses inferred actions for separate Procgen policy learning. It informs WAM representations without establishing a language-conditioned robotic WAM. The paper is retained for its adjacent resource/method role; WAM joint-prediction/IDM architecture quadrants do not apply to the cataloged contribution. Reading evidence
Contribution
FLAM learns a video world model whose slots each carry a latent action while sharing interaction-aware dynamics networks. Prediction-trained factorization improves rollouts supplied with future-inferred actions and can supply pseudo action labels for behavior cloning. Its strongest evidence concerns multi-entity video modeling; downstream control is evaluated separately in Procgen.
Abstract
An abstract has not been added yet.
Affiliations
University of Texas at Austin; Sony AI