Robotic task execution on Simulation Tasks Averaged (test)
87.5Success Rate (SR)AERMANI-VLM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AERMANI-VLMBackbone=Gemini-3-Pro, Manip.=✓2025.11 | 87.5 | 2.3 | 16.8 | 4.1 | 0.92 | |
| AERMANI-VLMBackbone=GPT-4o, Manip.=✓2025.11 | 85 | 2.6 | 18.4 | 5.2 | 0.88 | |
| AERMANI-VLMBackbone=Claude-3.5, Manip.=✓2025.11 | 82.5 | 3.1 | 19.6 | 6.4 | 0.95 | |
| AERMANI-VLMBackbone=Qwen2-VL, Manip.=✓2025.11 | 77.5 | 4.5 | 22.8 | 8.7 | 0.61 | |
| AERMANI-VLMBackbone=LLaVA-Next, Manip.=✓2025.11 | 72.5 | 5.2 | 25.3 | 10.1 | 0.58 | |
| UAV-VLAManip.=–2025.11 | 45 | 10.5 | 26 | 18 | 0.88 | |
| ReAct Prompting (VLM)Manip.=✓2025.11 | 35 | 12.5 | 28 | 15 | 0.9 | |
| UAV-VLNManip.=–2025.11 | 30 | 16.5 | 38 | 31 | 0.72 | |
| SayCanManip.=✓2025.11 | 20 | 16 | 33 | 28 | 0.91 | |
| RT-2Manip.=✓2025.11 | 10 | 19 | 42 | 35 | 0.85 |