Grounded Action Models Incorporate 3D Grounding for Robotics
September 19, 2026
Grounded Action Models (GAMs) use 3D grounding to condition manipulation policies on language, points, or box prompts. By converting prompts into object-centric representations of geometry and visual features, the model enables more precise metric grounding than implicit VLA models.
HOW THIS AFFECTS YOU
●
builderYou can implement more precise robot manipulation by providing explicit 3D spatial prompts.
●
researcherThis shifts the paradigm from implicit learning to explicit metric grounding in foundation models.