vision-language-action

Definition

A model class for robotics that takes images plus a language instruction and emits robot actions directly, often by discretizing actions into tokens and predicting them the way text is predicted. The usual recipe starts from a pre-trained vision-language model and adds an action head, inheriting general visual and linguistic grounding. What separates it from a VLM is the output: control signals executed in the physical world, where a mistake cannot be undone by resampling.

Explainers using this term

• Paper Explained: Beyond Data Scaling — Why the Backbone, Not the Trajectory Count, Decides Your VLA (VLAct)Beyond Data Scaling: Representation-Centric Continued Pre-training for Vision-Language-Action Models
• Paper Walkthrough: TurboVLA — Kick the LLM Out of the Loop and Run a Robot Policy at 32 Hz on an RTX 4090 with Under 1 GB of VRAMTurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM