Pairwise progress-judgment on RoboPulse Large hop range
79Accuracy (Real)VLAC
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VLACzero-shot=true2026.03 | 79 | 78 | 81 | 78 | 79 | |
| GVLzero-shot=true2026.03 | 78 | 83 | 78 | 75 | 78 | |
| Robo-Dopaminezero-shot=true2026.03 | 78 | 90 | 88 | 83 | 85 | |
| Gemini 3 Pro Previewzero-shot=true2026.03 | 72 | 85 | 77 | 74 | 77 | |
| Qwen3-VL-8B-Instructzero-shot=true, parameters=8B, variant=Instruct2026.03 | 68 | 79 | 77 | 69 | 74 | |
| CLIP ViT-L/14 (I2I)zero-shot=true, backbone=ViT-L/14, input_interface=I2I2026.03 | 64 | 72 | 68 | 54 | 65 | |
| CLIP ViT-B/32 (I2I)zero-shot=true, backbone=ViT-B/32, input_interface=I2I2026.03 | 60 | 69 | 63 | 51 | 61 | |
| Qwen3-VL-4B-Instructzero-shot=true, parameters=4B, variant=Instruct2026.03 | 60 | 71 | 65 | 68 | 66 | |
| GPT-5.2zero-shot=true2026.03 | 57 | 62 | 70 | 59 | 62 | |
| CLIP ViT-L/14 (T2I)zero-shot=true, backbone=ViT-L/14, input_interface=T2I2026.03 | 46 | 42 | 42 | 51 | 45 | |
| CLIP ViT-B/32 (T2I)zero-shot=true, backbone=ViT-B/32, input_interface=T2I2026.03 | 38 | 42 | 61 | 54 | 49 |