Continual Imitation Learning on LIBERO long 1.0
84.7Forward Transfer (FWT)TS-Stellar
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| TS-StellarModel Category=VLA Models, Training Strategy=Pretrained, Pretrained Data Source=cross-embodiment data2025.11 | 84.7 | 40.8 | 51.4 | 39.7 | |
| TS-StellarModel Category=CIL Methods, Training Strategy=Pretrained, Pretrained Data Source=LIBERO-902025.11 | 81.5 | 37.7 | 50.5 | 40.9 | |
| T-StellarModel Category=VLA Models, Training Strategy=Pretrained, Pretrained Data Source=cross-embodiment data2025.11 | 80.3 | 38 | 48.6 | 32.2 | |
| ERModel Category=CIL Methods, Training Strategy=Pretrained, Pretrained Data Source=LIBERO-902025.11 | 80.3 | 48.4 | 40.9 | 16.1 | |
| T-StellarModel Category=CIL Methods, Training Strategy=Pretrained, Pretrained Data Source=LIBERO-902025.11 | 79.3 | 36.8 | 48.3 | 36.3 | |
| T-StellarModel Category=VLA Models, Training Strategy=Scratch2025.11 | 76.3 | 41.6 | 41.2 | 34.2 | |
| T-StellarModel Category=CIL Methods, Training Strategy=Scratch2025.11 | 76.3 | 41.6 | 41.2 | 34.2 | |
| TS-StellarModel Category=VLA Models, Training Strategy=Scratch2025.11 | 75.5 | 37.3 | 43.6 | 35 | |
| TS-StellarModel Category=CIL Methods, Training Strategy=Scratch2025.11 | 75.5 | 37.3 | 43.6 | 35 | |
| MoDEModel Category=VLA Models, Training Strategy=Scratch2025.11 | 71.6 | 46.3 | 34.3 | 20.2 | |
| MoDEModel Category=VLA Models, Training Strategy=Pretrained, Pretrained Data Source=cross-embodiment data2025.11 | 71.1 | 46.4 | 32.9 | 19.9 | |
| ERModel Category=CIL Methods, Training Strategy=Scratch2025.11 | 70.8 | 47.3 | 31 | 18.9 | |
| UniVLAModel Category=VLA Models, Training Strategy=Pretrained, Pretrained Data Source=cross-embodiment data2025.11 | 70.3 | 31.3 | 43.6 | 17.1 | |
| IsCiLModel Category=CIL Methods, Training Strategy=Pretrained, Pretrained Data Source=LIBERO-902025.11 | 65.8 | 31.5 | 41.4 | 17.3 | |
| SeqLoRAModel Category=CIL Methods, Training Strategy=Pretrained, Pretrained Data Source=LIBERO-902025.11 | 62 | 56 | 16.6 | 6 | |
| π0Model Category=VLA Models, Training Strategy=Pretrained, Pretrained Data Source=cross-embodiment data2025.11 | 60.5 | 34.4 | 35.6 | 24.3 | |
| π0Model Category=VLA Models, Training Strategy=Scratch2025.11 | 57.3 | 48.7 | 22.2 | 12 | |
| π0.5Model Category=VLA Models, Training Strategy=Pretrained, Pretrained Data Source=cross-embodiment data2025.11 | 55.5 | 34.8 | 29.8 | 21.6 | |
| LoTUSModel Category=CIL Methods, Training Strategy=Pretrained, Pretrained Data Source=LIBERO-902025.11 | 32.3 | 2.8 | 29.4 | 31.8 | |
| IsCiLModel Category=CIL Methods, Training Strategy=Scratch2025.11 | 26 | 15 | 15.2 | 6.8 | |
| LoTUSModel Category=CIL Methods, Training Strategy=Scratch2025.11 | 11.5 | 0.8 | 11.3 | 10.5 | |
| SeqLoRAModel Category=CIL Methods, Training Strategy=Scratch2025.11 | 9.6 | 8.8 | 2.3 | 0.8 | |
| UniVLAModel Category=VLA Models, Training Strategy=Scratch2025.11 | 6.1 | 6 | 0.9 | 0 |