# OmniD: Generalizable Robot Manipulation Policy via Image-Based BEV Representation

arXiv: https://arxiv.org/abs/2508.11898 · HTML: https://mlsanity.com/paper/2508.11898/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 16 Aug 2025.

## Built on by (most cited first, 3 of 3)

- [VO-DP: Semantic-Geometric Adaptive Diffusion Policy for Vision-Only Robotic Manipulation](https://mlsanity.com/paper/2510.15530/) (2025)
- [RoboRouter: Training-Free Policy Routing for Robotic Manipulation](https://mlsanity.com/paper/2603.07892/) (2026)
- [Prepare Before You Act: Learning From Humans to Rearrange Initial States](https://mlsanity.com/paper/2509.18043/) (2025)

## Abstract (arXiv)

The visuomotor policy can easily overfit to its training datasets, such as fixed camera positions and backgrounds. This overfitting makes the policy perform well in the in-distribution scenarios but underperform in the out-of-distribution generalization. Additionally, the existing methods also have difficulty fusing multi-view information to generate an effective 3D representation. To tackle these issues, we propose Omni-Vision Diffusion Policy (OmniD), a multi-view fusion framework that synthesizes image observations into a unified bird's-eye view (BEV) representation. We introduce a deformable attention-based Omni-Feature Generator (OFG) to selectively abstract task-relevant features while suppressing view-specific noise and background distractions. OmniD achieves 11%, 17%, and 84% average improvement over the best baseline model for in-distribution, out-of-distribution, and few-shot experiments, respectively. Training code and simulation benchmark are available: https://github.com/1mather/omnid.git