Robotic Manipulation on LIBERO-V Visual
90.8Performance (Camera)π0.5 (One-Shot FLA)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| π0.5 (One-Shot FLA)Base Model=π0.5, Evaluation Protocol=One-Shot, Adaptation Method=Feature Linear Adaptation (FLA)2025.12 | 90.8 | 96.8 | 97.1 | 94.6 | 94.8 | |
| π0.5Base Model=π0.5, Evaluation Protocol=One-Shot, Adaptation Method=Lora Fine-tuned2025.12 | 90.3 | 96.5 | 97.2 | 94.5 | 94.6 | |
| π0.5 (One Shot FTM)Base Model=π0.5, Evaluation Protocol=One-Shot, Adaptation Method=Feature Token Modulation (FTM)2025.12 | 87.2 | 96 | 96 | 93.6 | 90.5 | |
| π0Base Model=π0, Evaluation Protocol=One-Shot, Adaptation Method=Lora Fine-tuned2025.12 | 83.6 | 90.7 | 86.7 | 87 | 87 | |
| OpenVLA-OFT-mBase Model=OpenVLA-OFT-m, Training data=Fine-Tuned on Libero-Plus2025.12 | 65.2 | 94.9 | 93.9 | 89.9 | 86 | |
| OpenVLA-OFTBase Model=OpenVLA-OFT2025.12 | 50.3 | 92.7 | 92.3 | 78.6 | 78.5 | |
| π0.5Base Model=π0.5, Evaluation Protocol=Zero-Shot2025.12 | 48.5 | 96.2 | 96 | 93.5 | 83.6 |