Multi-task Robotic Manipulation on LIBERO 90 tasks
97.3Easy Success RateDISC
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DISCEvaluation Protocol=trained-from-scratch, Visual Backbone=ResNet-18, Language Backbone=T5-small, Model Parameters=~30M2026.05 | 97.3 | — | 95.3 | 92.7 | 94.3 | |
| OTTEREvaluation Protocol=trained-from-scratch, Visual Backbone=ResNet-18, Language Backbone=T5-small, Model Parameters=~30M2026.05 | 96.4 | — | 87.9 | 83.2 | 86.6 | |
| π0.5Evaluation Protocol=Pretrained + FT2026.05 | 96.4 | — | 97 | 94.4 | 95.7 | |
| π0Evaluation Protocol=Pretrained + FT2026.05 | 95.9 | — | 92 | 90.2 | 91.6 | |
| DPEvaluation Protocol=trained-from-scratch, Visual Backbone=ResNet-18, Language Backbone=T5-small, Model Parameters=~30M2026.05 | 93.2 | — | 72.7 | 72.7 | 75.2 | |
| OctoEvaluation Protocol=trained-from-scratch, Visual Backbone=ResNet-18, Language Backbone=T5-small, Model Parameters=~30M2026.05 | 92.3 | — | 89 | 79.3 | 84.7 | |
| VQ-BeTEvaluation Protocol=trained-from-scratch, Visual Backbone=ResNet-18, Language Backbone=T5-small, Model Parameters=~30M2026.05 | 89.5 | — | 86.7 | 84.3 | 85.9 | |
| DiTEvaluation Protocol=trained-from-scratch, Visual Backbone=ResNet-18, Language Backbone=T5-small, Model Parameters=~30M2026.05 | 57.3 | — | 42 | 34.2 | 40.1 | |
| H-ZeroEvaluation Protocol=trained-from-scratch, Visual Backbone=ResNet-18, Language Backbone=T5-small, Model Parameters=~30M2026.05 | 50 | — | 75.1 | 69.1 | 69.1 | |
| H-GenEvaluation Protocol=trained-from-scratch, Visual Backbone=ResNet-18, Language Backbone=T5-small, Model Parameters=~30M2026.05 | 30.5 | — | 50.7 | 46.3 | 46.1 | |
| BAKUAction head=MLP2024.06 | — | 90 | — | — | — | |
| BAKUAction head=VQ-BeT2024.06 | — | 90 | — | — | — | |
| MT-ACT2024.06 | — | 54 | — | — | — | |
| RT-12024.06 | — | 16 | — | — | — |