Research map / Vision-language and multimodal / Robot action and navigation
Obstruction Reasoning for Robotic Grasping
Runyu Jiao, Matteo Bortolon, Francesco Giuliari, Alice Fasoli, Sergio Povoli, Guofeng Mei, Yiming Wang, Fabio Poiesi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add obstruction cue: occlusion ratio (ablation) | UNOGrasp SFT · UNOBench synthetic (Overall) | SR-F1: 74.7 -> 76.4 | Improved |
| add obstruction cue: contact point (ablation) | UNOGrasp SFT · UNOBench synthetic (Overall) | SR-F1: 74.7 -> 75.3 | Improved |
| add obstruction cue: degree word (ablation) | UNOGrasp SFT · UNOBench synthetic (Overall) | SR-F1: 74.7 -> 75.1 | No clear effect |
| add RFT with set-level IoU reward (ablation) | UNOGrasp · UNOBench synthetic (Overall) | SR-F1: 76.4 -> 78.2 | Improved |
| add RFT with IoU reward on hard scenes (ablation) | UNOGrasp · UNOBench synthetic (Hard) | SR-F1: 50.1 -> 54.5 | Improved |
| add reasoning chain and RFT vs SFT-only answer | UNOGrasp vs Qwen2.5-VL SFT · UNOBench synthetic Oracle (Hard) | SR-F1: 34.3 -> 54.5 | Improved |
| add reasoning chain and RFT vs SFT-only answer | UNOGrasp vs Qwen2.5-VL SFT · UNOBench real Oracle (Hard) | SR-F1: 25.9 -> 63.9 | Improved |
| in-context examples for Gemini (ablation) | Gemini Robotics-ER 1.5 · UNOBench synthetic Oracle (No obstruction) | SR: 68.7 -> 54.6 | Hurt |
| in-context examples for Gemini (ablation) | Gemini Robotics-ER 1.5 · UNOBench synthetic Oracle (Hard) | SR-F1: 5.4 -> 13.8 | Improved |
| UNOGrasp vs Gemini in real robot trials | UNOGrasp · Real robot UR5e (Hard) | success rate: 10% -> 40% | Improved |
| UNOGrasp vs Gemini in real robot trials | UNOGrasp · Real robot UR5e (Easy and Medium) | success rate: 80% vs 80%, 30% vs 30% | No clear effect |
| UNOGrasp NL prompting vs Oracle SoM grounding | UNOGrasp · UNOBench synthetic (Hard) | SR-F1: 54.5 -> 37.2 | Hurt |
Closest papers by meaning
- RealVLG-R1: A Large-Scale Real-World Visual-Language Grounding Benchmark for Robotic Perception and Manipulation (CVPR 2026 · similarity 0.88)
- AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation (ICLR 2025 · similarity 0.88)
- PAC Bench: Do Foundation Models Understand Prerequisites for Executing Manipulation Policies? (NeurIPS 2025 · similarity 0.87)
- OmniManip: Towards General Robotic Manipulation via Object-Centric Interaction Primitives as Spatial Constraints (CVPR 2025 · similarity 0.87)
- RoboProcessBench: Benchmarking Process-Aware Understanding in Visual-Language Robotic Manipulation (NeurIPS 2026 · similarity 0.87)
- RoboMamba: Efficient Vision-Language-Action Model for Robotic Reasoning and Manipulation (NeurIPS 2024 · similarity 0.86)
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.
