Path-level reasoning on UNOBench synthetic Easy (test)
82.8SR (Precision)UNOGrasp
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| UNOGraspPrompting Type=Oracle (with SoM)2025.11 | 82.8 | 84.4 | 83.3 | 0.11 | |
| UNOGraspPrompting Type=Natural Language Prompting2025.11 | 74.8 | 75.1 | 74.9 | 0.2 | |
| Qwen2.5-VLPrompting Type=Oracle (with SoM), Evaluation Protocol=SFT2025.11 | 69.6 | 70.3 | 69.8 | — | |
| Gemini Robotics-ER 1.5Prompting Type=Oracle (with SoM), Evaluation Protocol=ICL2025.11 | 67.1 | 73.3 | 69.1 | 0.2 | |
| Qwen2.5-VLPrompting Type=Natural Language Prompting, Evaluation Protocol=SFT2025.11 | 65.2 | 65.5 | 65.3 | — | |
| Gemini Robotics-ER 1.5Prompting Type=Natural Language Prompting, Evaluation Protocol=ICL2025.11 | 60.6 | 61.8 | 61 | 0.8 | |
| Gemini Robotics-ER 1.5Prompting Type=Oracle (with SoM)2025.11 | 57.7 | 62.8 | 59.3 | 0.25 | |
| Gemini Robotics-ER 1.5Prompting Type=Natural Language Prompting2025.11 | 51.8 | 52.8 | 52.1 | 0.84 | |
| Qwen2.5-VLPrompting Type=Oracle (with SoM), Evaluation Protocol=ICL2025.11 | 24.7 | 36.7 | 27.2 | 0.55 | |
| Qwen2.5-VLPrompting Type=Natural Language Prompting, Evaluation Protocol=ICL2025.11 | 11.8 | 13.1 | 12.2 | 0.86 |