Open Models Almanac
Methodsmodel architecturemultimodal architecture

specific method · filed under model architecture

Separate flow-matching action expert

A separate flow-matching action expert is a smaller transformer, alongside a VLM backbone, that decodes continuous robot actions by denoising noisy action tokens with a flow-matching objective.1 It receives VLM information only unidirectionally through self-attention, and its outputs are projected into the target vector field.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. The action expert is a smaller transformer that takes in a sequence of noisy action tokens for an action horizon of 50, i.e. , and is trained with the flow matching objective — π0.5\pi_{0.5}: a Vision-Language-Action Model with Open-World Generalization
  2. In effect, information flows unidirectionally from the VLM to the action expert; no VLM embedding attends to the action expert — π0.5\pi_{0.5}: a Vision-Language-Action Model with Open-World Generalization

Also called action expert.

sources
3
models
2
labs adopt it
2
strongest
core

Documented in

Physical Intelligence

Evidence

3 quotes from 3 sources, one card per model, each opening with its strongest.

How sources treat it
core 3

π0.5

Physical Intelligenceadopted2 sources

we adapt the model to also have an action expert, as with π0, in order to both represent actions with finer granularity and enable more compute-efficient inference for real-time control

coreunclearin π0.5
1 more from this model

we use a novel action expert that augments the standard VLM with flow-based outputs.

coreunclearin π0Physical Intelligence
↑