Robotic Manipulation on Real-world Manipulation SO-100
60Place Success RateRoboFAC-7B (Low)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RoboFAC-7B (Low)Latency (s)=6.7 ± 0.5, Attempts=5 attempts2025.05 | 60 | 85 | 30 | 70 | 61.25 | |
| GPT-4oLatency (s)=24.3 ± 3.4, Attempts=5 attempts2025.05 | 50 | 80 | 30 | 65 | 56.25 | |
| Qwen2.5-VL-7BLatency (s)=6.9 ± 0.6, Attempts=5 attempts2025.05 | 50 | 70 | 20 | 60 | 50 | |
| RoboFAC-7B (High)Latency (s)=7.0 ± 0.5, Attempts=5 attempts2025.05 | 50 | 75 | 20 | 55 | 50 | |
| RoboFAC-7B (High)Latency (s)=7.0 ± 0.5, Attempts=1 attempt2025.05 | 45 | 65 | 10 | 45 | 41.25 | |
| No correctionAttempts=5 attempts2025.05 | 40 | 70 | 20 | 60 | 47.5 | |
| RoboFAC-7B (Low)Latency (s)=6.7 ± 0.5, Attempts=1 attempt2025.05 | 40 | 70 | 20 | 50 | 45 | |
| Qwen2.5-VL-7BLatency (s)=6.9 ± 0.6, Attempts=1 attempt2025.05 | 35 | 60 | 15 | 45 | 38.75 | |
| GPT-4oLatency (s)=24.3 ± 3.4, Attempts=1 attempt2025.05 | 25 | 70 | 15 | 50 | 40 | |
| No correctionAttempts=1 attempt2025.05 | 20 | 55 | 10 | 35 | 30 |