Robotic Manipulation on VIMA-Bench
99.3Task 1 ScoreVIMA-20M
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VIMA-20MMethod Type=Learning-based2026.04 | 99.3 | 100 | 100 | 99.5 | 58.7 | 41.5 | 83.2 | |
| EEAgentBase LLM=ChatGPT-4o, Segmentation Model=Huge SAM, Max Trials=3, LSTRO=True2026.04 | 94.3 | 94.2 | 99.3 | 87.2 | 84.8 | 93.2 | 92.2 | |
| Instruct2ActMethod Type=LLM Planning2026.04 | 91.8 | 80.1 | 98.4 | 78.2 | 72.7 | 89.6 | 85.1 | |
| CoT-PromptMethod Type=Prompting Strategies2026.04 | 91.7 | 92.8 | 99.2 | 82.5 | 80 | 88.3 | 88.2 | |
| S-C (Select)Method Type=Prompting Strategies2026.04 | 91.2 | 93.7 | 99.2 | 82.8 | 80.5 | 88.3 | 89.3 | |
| S-C (Reflect)Method Type=Prompting Strategies2026.04 | 90.8 | 92.5 | 99 | 82.7 | 79.8 | 87.2 | 88.7 | |
| S-C (Voting)Method Type=Prompting Strategies2026.04 | 90.5 | 93 | 98.8 | 83.3 | 81.2 | 87.7 | 89.2 | |
| Vannila EAMethod Type=Prompting Strategies, LSTRO=False2026.04 | 89.2 | 90.8 | 98.3 | 76.2 | 74.5 | 81.8 | 85.1 | |
| CLINMethod Type=LLM Planning2026.04 | 81.7 | 77.7 | 80.7 | 72.8 | 70.3 | 72.7 | 76 | |
| CaPMethod Type=LLM Planning2026.04 | 76.3 | 73.7 | 79.3 | 70.8 | 66.3 | 60.3 | 71.1 | |
| VIMA-GPTMethod Type=Learning-based2026.04 | 53 | 57.2 | 50.52 | 46.3 | 37.2 | 1.7 | 47.6 | |
| VIMA-FlamingoMethod Type=Learning-based2026.04 | 52.8 | 55 | 54.5 | 51 | 40.3 | 0.5 | 49.2 | |
| VIMA-GatoMethod Type=Learning-based2026.04 | 50.7 | 56.2 | 38.7 | 52.3 | 36 | 1 | 45.9 |