Robotic Manipulation on LIBERO Franka Panda 120 Tasks
98.8Spatial Achievement Rateπ0.5
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| π0.5Size=3B, Training/Fine-tuning Protocol=substantial LIBERO fine-tuning (30k+ demos), Evaluation Mode=fine-tuned, Notes=LIBERO-10 subset2026.01 | 98.8 | 98.2 | 98 | 92.4 | 96.85 | |
| SmolVLASize=2.25B, Training/Fine-tuning Protocol=pre-trained, minimal LIBERO fine-tuning2026.01 | 93 | 94 | 91 | 77 | 88.75 | |
| FAEA Baseline (no limit)Size=API, Training/Fine-tuning Protocol=demonstration-free, Observation/State Type=privileged state, Evaluation Mode=Baseline (no limit)2026.01 | 90 | 100 | 80 | 83.3 | 84.9 | |
| FAEA + coachingSize=API, Training/Fine-tuning Protocol=demonstration-free, Observation/State Type=privileged state, Evaluation Mode=+ coaching2026.01 | 90 | 100 | 90 | 86.7 | 88.2 | |
| SmolVLASize=0.45B, Training/Fine-tuning Protocol=pre-trained, minimal LIBERO fine-tuning2026.01 | 90 | 96 | 92 | 71 | 87.3 | |
| π0Size=3.3B, Training/Fine-tuning Protocol=pre-trained, minimal LIBERO fine-tuning, Evaluation Mode=pretrained2026.01 | 90 | 86 | 95 | 73 | 86 | |
| OpenVLASize=7B, Training/Fine-tuning Protocol=pre-trained, minimal LIBERO fine-tuning2026.01 | 84.7 | 88.4 | 79.2 | 53.7 | 76.5 | |
| Diffusion PolicyTraining/Fine-tuning Protocol=pre-trained, minimal LIBERO fine-tuning2026.01 | 78.3 | 92.5 | 68.3 | 50.5 | 72.4 | |
| FAEA Pilot (10 trials)Size=API, Training/Fine-tuning Protocol=demonstration-free, Observation/State Type=privileged state, Evaluation Mode=Pilot (10 trials)2026.01 | 70 | 100 | 60 | 68.9 | 70.6 |