SciGroveBeta
Robotics

Geometric Action Model for Robot Policy Learning

Jisang Han, Seonghu Jeon, Jaewoo Jung, et al.

Featured July 12, 2026

Original Paper

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

A new robot brain called GAM uses a special 3D vision model to predict both what the world will look like next and what actions the robot should take, making it much better at handling tricky situations and moving faster.

In depth
The paper introduces the Geometric Action Model (GAM), a novel robot policy that directly repurposes a pretrained Geometric Foundation Model (GFM). This allows for joint prediction of future 3D geometry and robot actions within a single shared backbone, enhancing robustness and efficiency by explicitly leveraging geometric priors for language-conditioned manipulation tasks.

Key Takeaways

  • 1
    GFM Repurposing: The authors directly integrate a pretrained Geometric Foundation Model (GFM) as a shared backbone for perception, temporal prediction, and action decoding, moving beyond static feature extraction.
  • 2
    Joint Prediction: GAM simultaneously forecasts future 3D geometry and robot action chunks within a single autoregressive token sequence, enabling tight interaction between spatial representations and actions.
  • 3
    Enhanced Robustness & Efficiency: The proposed model achieves superior accuracy and out-of-distribution robustness (especially to camera perturbations) while being significantly faster and lighter than existing vision-language-action (VLA) and world-action model (WAM) baselines.

Conceptual Flow

HIGH LEVEL
1
Unifying 3D Vision and Robot Actions

The robot's brain uses a special 3D vision model to see, predict what will happen next in 3D, and decide what to do, all in one go.

Robot Sees
User Tells
Think in 3D
Future 3D World
Robot Moves
2
Better, Faster, Stronger Robot Control

This new method makes robots much better at difficult tasks, especially when cameras move, and they also work faster and are smaller.

Old Robot Brains
New 3D Brain
Higher Success
Faster Decisions
Handles Changes