G^3VLA: Geometric Inductive Bias for Vision-Language-Action Models

Key points

• We propose G^3VLA, a camera-aware geometric module that injects calibrated structure into the visual-token stream of a pretrained VLA without altering its action space or original action-learning objective.1
• G^3VLA combines intrinsic-conditioned ray embeddings, projective positional encoding (PRoPE), and bidirectional cross-view fusion to establish geometrically consistent representations across camera views.1
• On π0, G^3VLA consistently improves performance on LIBERO and RoboCasa24 under both supervision regimes, while ground-truth supervision improves five-task RoboTwin2.0 averages with three cameras.1
• These results demonstrate that grounding VLA visual tokens in calibrated camera geometry provides a simple and general mechanism for improving geometric consistency, multi-view reasoning, and viewpoint robustness in pretrained robot policies.1

Sources (1)

• [1]G^3VLA: Geometric Inductive Bias for Vision-Language-Action Models
  Hugging Face Daily Papers · Oct 14, 12:00 AM

  Vision-language-action (VLA) models have rapidly advanced generalist robot manipulation by harnessing the semantic knowledge of pretrained vision-language backbones.

  We propose G^3VLA, a camera-aware geometric module that injects calibrated structure into the visual-token stream of a pretrained VLA without altering its action space or original action-learning objective.

Extractive summary: sentences quoted from the sources.