AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

UniRAP: Towards Unified Part-level Physical Affordance Reasoning and Actionable Perception

Linfei Li, Ruining Hu, Lin Zhang, Zhong Wang, Fengyi Zhang, Ying Shen, Binqiang Wang, Xin Zhang

NeurIPS 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes UniRAP, a unified model that infers part-level physical affordances from visual observations and maps language instructions to actionable geometric representations.

Problem

Grounding high-level language semantics into part-level physical interaction (affordance reasoning) remains challenging for vision-language perception models.

Approach

A single unified model that performs part-level physical affordance reasoning and maps language instructions to actionable geometric representations, rather than separate perception and grounding stages.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add affordance pre-training stage (ablation)UniRAP 3B · 3DOI referringgIoU: 40.5 -> 50.4Improved
add affordance pre-training stage (ablation)UniRAP 3B · 3DOI reasoninggIoU: 37.5 -> 52.9Improved
add actionable affordance fine-tuning (ablation)UniRAP 3B · RealVLG-Benchgrasp gAcc: 19.6 -> 63.3Improved
add visual referring pre-training (ablation)UniRAP 3B · RealVLG-Benchgrasp gAcc: 26.4 -> 63.3Improved
use regression-based Grasp Head instead of heatmap head (ablation)UniRAP 3B · RealVLG-Benchgrasp gAcc: 13.4 -> 63.3Improved
add Bbox Head (ablation)UniRAP 3B · ReasonSeggIoU: 56.8 -> 57.9Improved
add Bbox Head (ablation)UniRAP 3B · ReVOSJ&F: 60.4 -> 61.9Improved
add Bbox Head (ablation)UniRAP 3B · RealVLG-Benchgrasp gAcc: 24.7 -> 63.3Improved
heatmap grasp head instead of UniRAP head (ablation)UniRAP 3B · 3DOI referringgIoU: 52.3 vs 52.1No clear effect
use UniRAP (3B) instead of RAGNet (7B)affordance segmentation · 3DOI zero-shot reasoninggIoU: +17.6Improved
end-to-end grasp head instead of text-generated graspUniRAP vs RealVLG-R1 · RealVLG-Bench Novelgrasp gAcc: 18.9 -> 63.3Improved
UniRAP 3B vs larger specialized VideoLISA on reasoningUniRAP · ReasonSegcIoU: 67.1 (VideoLISA) vs 59.0Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy