Paper page - ACE-Brain-0.6: Scaling Spatial Grounding for Physical Foundation Models

Source: https://huggingface.co/papers/2610.27040 Authors:

,

,

,

,

,

,

,

,

,

,

Abstract

Scaling diverse verifiable visual grounding data improves spatial perception and embodied manipulation performance in open multimodal models.

General-purposemultimodal models, including GPT-6 Astra and Claude Opus 5.5, demonstrate broad capabilities inembodied manipulation. Our analysis indicates that this generality is accompanied by strongspatial perception, as Astra can determine where to interact with task-relevant entities amid visual distractors. Becausespatial perceptionconnects high-level reasoning to the physical scene, we investigate scalingvisual groundingto strengthen the perceptual foundation of general embodied operation. We develop a unified data pipeline that combinesrobot manipulationrecordings with complementary visual data from diverse sources. After automatically annotating the robot recordings, the pipeline converts the collected data into verifiablegrounding supervision.Fine-tuningQwen3.5-9Bon approximately 1.5 million samples produced by this pipeline yields ACE-Brain-0.6. The model outperforms GPT-6 Astra on four of seven public grounding benchmarks, with an overall average score of 75.41 compared with Astra’s 73.68. Under a fixedQwen3.5-9Bfine-tuningsetup, aggregate grounding performance improves steadily as successive data releases expand the training collection to cover a broader range of grounding scenarios. These results show that scaling verifiablegrounding supervisionfrom diverse sources can strengthen thespatial perceptionof an open 9B model for general embodied operation.

View arXiv pageView PDFProject pageGitHub8Add to collection

Get this paper in your agent:

hf papers read 2610\.27040

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper1

#### ACERobotics/ACE-Brain-0.6-9B Image-Text-to-Text• 9B• Updated1 day ago • 38 • 2

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.27040 in a dataset README.md to link it from this page.

Spaces citing this paper1

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.