Spatial Reasoning on RoboInter-VQA Spatial
10.9Grasp Pose ErrorGPT4o-mini
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| GPT4o-miniSize=mini2026.02 | 10.9 | 6.8 | — | 7.2 | 1.1 | 16.8 | 53.6 | |
| QwenVL2.5-3BParameters=3B, Base Model=Qwen2.5-VL2026.02 | 21.1 | 46.6 | 12.2 | 34.1 | 6.1 | 21.9 | 50.9 | |
| Robobrain-2.0-7BParameters=7B2026.02 | 23.3 | — | — | — | 2.5 | 21.5 | 49.2 | |
| Llava-OV-7BParameters=7B, Base Model=LLaVA-OneVision2026.02 | 24.5 | 25.8 | 5.5 | 23.7 | 1.6 | 31.9 | 54.4 | |
| InternVL3-1BParameters=1B2026.02 | 24.8 | 7.8 | 2.3 | 8.3 | 1.2 | 25.9 | 50.4 | |
| InternVL3-8BParameters=8B2026.02 | 25.1 | 32.7 | 5.9 | 28.2 | 3.5 | 31.1 | 52.9 | |
| InternVL3-2BParameters=2B2026.02 | 25.5 | 20.6 | 3.1 | 17.9 | 1.9 | 27.3 | 50.1 | |
| Robobrain-2.0-3BParameters=3B2026.02 | 25.5 | 15.2 | — | — | 2.8 | 26.5 | 50.4 | |
| QwenVL2.5-7BParameters=7B, Base Model=Qwen2.5-VL2026.02 | 27.3 | 51.2 | 14.7 | 38.2 | 10.2 | 25.7 | 52.5 | |
| Gemini-2.5-flashSize=flash2026.02 | 32.7 | 1.7 | — | 1.2 | — | 69.4 | 65.5 | |
| RoboInter-Qwen-3BParameters=3B, Backbone=Qwen2026.02 | 74 | 76.1 | 34.9 | 52.7 | 61.6 | 73.4 | 75.2 | |
| RoboInter-LlavaOV-7BParameters=7B, Backbone=LLaVA-OneVision2026.02 | 74.1 | 82.9 | 46.3 | 55.1 | 70.1 | 79.7 | 76.3 | |
| RoboInter-Qwen-7BParameters=7B, Backbone=Qwen2026.02 | 76.1 | 75.1 | 37.8 | 56.9 | 62 | 75.7 | 75.6 |