Predict robot actions
Produce a robot-control action from visual observations and an instruction.
Desired outcome
A candidate action for a specified robot embodiment.
Explore from the outcome
Where this function can lead
These documented connections are starting points for selection, not a ranked list.
Model releases
Identifiable models documented for this function.
- OpenVLA 7B
Predicts robot manipulation actions from a camera image and an English-language instruction.
Link evidence: OpenVLA 7B model card
Typical inputs
- Camera image, language instruction, and robot context
Typical outputs
- Normalized or platform-specific action values
Sources
Last reviewed 2026-09-30