Multi-Task Learning on LIBERO
99.2Object ScoreNORA-Long
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NORA-LongTraining Strategy=RFT (Ours), Action Model Category=Discrete Action Models2026.02 | 99.2 | 98.2 | 95.8 | 89 | 95.6 | |
| π0Training Strategy=SFT, Action Model Category=Continuous Action Models2026.02 | 98.8 | 96.8 | 95.8 | 85.2 | 94.2 | |
| OpenVLA-OFTTraining Strategy=SFT, Action Model Category=Continuous Action Models2026.02 | 98.1 | 96.9 | 95.5 | 91.1 | 95.4 | |
| GR00T N1Training Strategy=SFT, Action Model Category=Continuous Action Models2026.02 | 97.6 | 94.4 | 93 | 90.6 | 93.9 | |
| NORA-Long (Baseline)Training Strategy=SFT, Action Model Category=Discrete Action Models2026.02 | 97.5 | 96.4 | 91 | 82.4 | 91.8 | |
| π0-FastTraining Strategy=SFT, Action Model Category=Discrete Action Models2026.02 | 96.8 | 96.4 | 88.6 | 60.2 | 85.5 | |
| NORA-1.5Training Strategy=SFT, Action Model Category=Continuous Action Models2026.02 | 96.4 | 97.3 | 94.5 | 89.6 | 94.5 | |
| WorldVLATraining Strategy=SFT, Action Model Category=Discrete Action Models2026.02 | 96.2 | 87.6 | 83.4 | 60 | 79.1 | |
| NORA-1.5 (DPO)Training Strategy=SFT + RFT, Action Model Category=Continuous Action Models2026.02 | 96 | 98 | 95.4 | 90.5 | 95 | |
| MolmoAct-7B-DTraining Strategy=SFT, Action Model Category=Discrete Action Models2026.02 | 95.4 | 87 | 87.6 | 77.2 | 86.6 | |
| VLA-RFTTraining Strategy=SFT + RFT, Action Model Category=Continuous Action Models2026.02 | 94.4 | 94.4 | 95.4 | 80.2 | 91.1 | |
| LiMoDEMulti-task learning=true2026.06 | 94.2 | 90.2 | 89.2 | 81.4 | 88.7 | |
| UniVLAMulti-task learning=true, Pre-training (LIBERO-90)=re-implementation2026.06 | 93.8 | 92.6 | 86.6 | 63 | 84 | |
| DPMulti-task learning=true2026.06 | 92.5 | 78.3 | 68.3 | 50.5 | 72.4 | |
| TGRPOTraining Strategy=SFT + RFT, Action Model Category=Discrete Action Models2026.02 | 92.2 | 90.4 | 81 | 59.2 | 80.7 | |
| CoT-VLATraining Strategy=SFT, Action Model Category=Discrete Action Models2026.02 | 91.6 | 87.5 | 87.6 | 69 | 83.9 | |
| ThinkActTraining Strategy=SFT + RFT, Action Model Category=Continuous Action Models2026.02 | 91.4 | 88.3 | 87.1 | 70.9 | 84.4 | |
| SpatialVLATraining Strategy=SFT, Action Model Category=Discrete Action Models2026.02 | 89.9 | 88.2 | 78.6 | 55.5 | 78.1 | |
| OpenVLATraining Strategy=SFT, Action Model Category=Discrete Action Models2026.02 | 88.4 | 84.7 | 79.2 | 53.7 | 76.5 | |
| OpenVLAMulti-task learning=true2026.06 | 88.4 | 84.7 | 79.2 | 53.7 | 76.5 | |
| SDPMulti-task learning=true2026.06 | 87.5 | 78.5 | 73.5 | 64.8 | 75.1 | |
| PPLMulti-task learning=true, Pre-training (LIBERO-90)=re-implementation2026.06 | 86 | 85 | 86 | 80 | 84 | |
| Octo-BaseTraining Strategy=SFT, Action Model Category=Continuous Action Models2026.02 | 85.7 | 78.9 | 84.6 | 51.1 | 75.1 | |
| TraceVLATraining Strategy=SFT, Action Model Category=Discrete Action Models2026.02 | 85.2 | 84.6 | 75.1 | 54.1 | 74.8 | |
| UniActionMulti-task learning=true2026.06 | 78 | 65 | 68 | 47 | 64.5 |