Robotic Manipulation on LIBERO (Success Rates)
99.2Long-horizon Success RatedVLA-RL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| dVLA-RLMethod Category=VLA-Based2026.06 | 99.2 | 99.8 | 100 | 99.6 | 99.7 | |
| SimpleVLA-RLMethod Category=VLA-Based2026.06 | 98.5 | 99.4 | 99.1 | 99.2 | 99.1 | |
| LingBot-VAMethod Category=WAM-Based2026.06 | 98.5 | 98.5 | 99.6 | 97.2 | 98.5 | |
| SimpleVLA-RLReward=Env. Success2026.06 | 98.5 | 99.4 | 99.1 | 99.2 | 99.1 | |
| AVA-VLAPolicy Training Protocol=One policy for all 4 suites2026.06 | 98.1 | 97.8 | 99.4 | 97.8 | 98.3 | |
| X-VLAArchitecture=Standard VLA2026.06 | 97.6 | 98.2 | 98.6 | 97.8 | 98.1 | |
| MotusMethod Category=WAM-Based2026.06 | 97.6 | 96.8 | 99.8 | 96.6 | 97.7 | |
| Cosmos PolicyMethod Category=WAM-Based2026.06 | 97.6 | 98.1 | 100 | 98.2 | 98.5 | |
| TTP2026.07 | 97 | 98.8 | 98.2 | 98.2 | 98.1 | |
| ACoT-VLAGuidance=Action2026.07 | 97 | 98.6 | 99 | 99.4 | 98.5 | |
| BeingH-0.52026.07 | 96.6 | 98.8 | 97.6 | 98.8 | 98 | |
| TTP w/o pre-training2026.07 | 96.6 | 98 | 97.4 | 97.6 | 97.4 | |
| AVA-VLAPolicy Training Protocol=One policy per suite2026.06 | 96.5 | 99.6 | 99.7 | 98.7 | 98.6 | |
| S2-VLAScale (B)=22026.06 | 96.4 | 98.4 | 99.6 | 98.4 | 98.2 | |
| UniFSArchitecture=Ours2026.06 | 95.6 | 99.6 | 99.6 | 98.2 | 98.3 | |
| VLA-CacheArchitecture=Fast-Slow VLA2026.06 | 95.4 | 98.3 | 97.5 | 98.3 | 97.4 | |
| CAC-VLAGuidance=Action2026.07 | 95.4 | 98.4 | 99.8 | 99.6 | 98.3 | |
| OpenVLA-OFTPolicy Training Protocol=One policy for all 4 suites2026.06 | 95.3 | 97.7 | 98 | 96.1 | 96.8 | |
| T2VLA (OpenVLA-OFT)Reward=Self-Reward2026.06 | 95.3 | 97.7 | 99.6 | 96.1 | 97.2 | |
| MIRTH2026.06 | 95.3 | 98.2 | 100 | 98.8 | 98.1 | |
| CogVLAArchitecture=Enhanced VLA2026.06 | 95.2 | 98.5 | 98.8 | 96.5 | 97.4 | |
| Fast-WAMMethod Category=WAM-Based2026.06 | 95.2 | 98.2 | 100 | 97 | 97.6 | |
| VLA-AdapterPolicy Training Protocol=One policy per suite2026.06 | 95 | 97.8 | 99.2 | 97.2 | 97.3 | |
| VLA-AdapterGuidance=Linguistics2026.07 | 95 | 97.8 | 99.2 | 97.2 | 97.3 | |
| FLOWERPolicy Training Protocol=One policy per suite2026.06 | 94.9 | 97.5 | 99.1 | 96.1 | 96.9 | |
| EO-1Architecture=Standard VLA2026.06 | 94.8 | 99.7 | 99.8 | 99.2 | 98.2 | |
| OpenVLA-OFTPolicy Training Protocol=One policy per suite2026.06 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | |
| OpenVLA-OFTSpeedup=1.00×2026.06 | 94.5 | 97.6 | 96.5 | 97.9 | 96.6 | |
| OpenVLA-OFTArchitecture=Enhanced VLA2026.06 | 94.5 | 96.2 | 98.3 | 96.2 | 95.3 | |
| OpenVLA-OFTMethod Category=VLA-Based2026.06 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | |
| OpenVLA-OFTScale (B)=72026.06 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | |
| OpenVLA-OFTvariant=OFT2026.06 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | |
| OpenVLA-OFT2026.07 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | |
| OpenVLA-OFTGuidance=Linguistics2026.07 | 94.5 | 97.6 | 98.4 | 97.9 | 97.1 | |
| EVOLVE-VLAReward=Learned Critic2026.06 | 94.4 | 95.4 | 97.4 | 95.8 | 95.8 | |
| GE-ActGuidance=Visual2026.07 | 94.4 | 98.2 | 97.6 | 95.8 | 96.5 | |
| UnifiedVLAPolicy Training Protocol=One policy for all 4 suites2026.06 | 94 | 95.4 | 98.8 | 93.6 | 95.5 | |
| UnifiedVLAScale (B)=8.52026.06 | 94 | 95.4 | 98.8 | 93.6 | 95.5 | |
| CronusVLAScale (B)=72026.06 | 94 | 97.3 | 99.6 | 96.9 | 97 | |
| UniVLAGuidance=Visual2026.07 | 94 | 95.4 | 98.8 | 93.6 | 95.5 | |
| RIPT-VLAPolicy Training Protocol=One policy per suite2026.06 | 93.8 | 99 | 98.6 | 98.6 | 97.5 | |
| QWen3+Flow matching+adapterInference Step=12026.06 | 93.6 | 99 | 99.6 | 97.6 | 97.4 | |
| VLA-AdapterArchitecture=Ours2026.06 | 93.6 | 97.2 | 98.8 | 93.4 | 95.8 | |
| MemoryVLAArchitecture=Enhanced VLA2026.06 | 93.4 | 98.4 | 98.4 | 96.4 | 96.5 | |
| MemoryVLAScale (B)=72026.06 | 93.4 | 98.4 | 98.4 | 96.4 | 96.7 | |
| MemoryVLAGuidance=Linguistics2026.07 | 93.4 | 98.4 | 98.4 | 96.4 | 96.7 | |
| OpenVLA-OFT_m2026.07 | 93.2 | 95.2 | 94.2 | 95.2 | 94.5 | |
| MM-ACTMethod Category=VLA-Based2026.06 | 93 | 97.8 | 99.4 | 94.8 | 96.3 | |
| InternVLA-M12026.07 | 92.6 | 98 | 99 | 93.8 | 95.9 | |
| π0.5Method Category=VLA-Based2026.06 | 92.4 | 98.8 | 98.2 | 98 | 96.9 | |
| π0.52026.07 | 92.4 | 98.8 | 98.2 | 98 | 96.8 | |
| π0.5Guidance=Linguistics2026.07 | 92.4 | 98.8 | 98.2 | 98 | 96.9 | |
| UniVLAPolicy Training Protocol=One policy per suite2026.06 | 92 | 96.5 | 96.8 | 95.6 | 95.2 | |
| QWen3+GROOTInference Step=42026.06 | 92 | 97.8 | 98.8 | 97.4 | 96.5 | |
| UniVLAMethod Category=VLA-Based2026.06 | 92 | 96.5 | 96.8 | 95.6 | 95.2 | |
| UniVLAScale (B)=72026.06 | 92 | 96.5 | 96.8 | 95.6 | 95.2 | |
| UniVLAReward=None (SFT)2026.06 | 92 | 96.5 | 96.8 | 95.6 | 95.2 | |
| UniVLA2026.07 | 92 | 96.5 | 96.8 | 95.6 | 95.2 | |
| Discrete Diffusion VLA2026.07 | 92 | 97.2 | 98.6 | 97.4 | 96.3 | |
| DD-VLAGuidance=Linguistics2026.07 | 92 | 97.2 | 98.6 | 97.4 | 96.3 | |
| PD-VLAPolicy Training Protocol=One policy per suite2026.06 | 91.7 | 95.5 | 96.7 | 94.9 | 94.7 | |
| PD-VLAArchitecture=Fast-Slow VLA2026.06 | 91.7 | 95.5 | 96.7 | 94.9 | 94.7 | |
| F12026.07 | 91.3 | 98.2 | 97.8 | 95.4 | 95.7 | |
| F1Guidance=Visual2026.07 | 91.3 | 98.2 | 97.8 | 95.4 | 95.7 | |
| GR00T-N1.5Speedup=1.00×2026.06 | 91 | 90.8 | 98.4 | 95.4 | 93.9 | |
| OpenVLA-OFT_w2026.07 | 90.7 | 96.2 | 98.3 | 96.2 | 95.3 | |
| GR00T N1Architecture=Fast-Slow VLA2026.06 | 90.6 | 94.4 | 97.6 | 93 | 93.9 | |
| GR00T N1Scale (B)=22026.06 | 90.6 | 94.4 | 97.6 | 93 | 93.9 | |
| GR00T-N12026.07 | 90.6 | 94.4 | 97.6 | 93 | 93.9 | |
| GR00T-N1Guidance=Linguistics2026.07 | 90.6 | 94.4 | 97.6 | 93 | 93.9 | |
| π0Speedup=1.00×2026.06 | 90 | 94.6 | 98.2 | 95.4 | 94.6 | |
| DreamVLAGuidance=Visual2026.07 | 89.5 | 97.5 | 94 | 89.5 | 92.6 | |
| GO-1Guidance=Linguistics2026.07 | 89.2 | 96.2 | 97.8 | 96 | 94.8 | |
| FastVSpeedup=1.44×2026.06 | 88.8 | 94.6 | 95.8 | 94 | 93.3 | |
| MM-ACT*Method Category=VLA-Based2026.06 | 88.7 | 91.2 | 82.6 | 90 | 88.1 | |
| GR00T-N1.5-CLPSpeedup=1.42×2026.06 | 88.6 | 89.4 | 98.8 | 95.8 | 93 | |
| QWen3+PIInference Step=42026.06 | 88.4 | 98.8 | 99.6 | 95.8 | 95.7 | |
| π02026.07 | 88.4 | 98 | 96.8 | 94.4 | 94.4 | |
| RIPT-VLA2026.07 | 87.5 | 92.7 | 95.6 | 98.4 | 93.6 | |
| OpenVLA-OFTReward=None (SFT)2026.06 | 86.5 | 91.6 | 95.3 | 90.6 | 91 | |
| π0-CLPSpeedup=1.39×2026.06 | 86.4 | 95 | 99.2 | 95 | 93.9 | |
| π0.5Architecture=Standard VLA2026.06 | 85.8 | 98 | 97.8 | 95.6 | 94.3 | |
| π0Policy Training Protocol=One policy for all 4 suites2026.06 | 85.2 | 96.8 | 98.8 | 95.8 | 94.2 | |
| π0Architecture=Standard VLA2026.06 | 85.2 | 96.8 | 98.8 | 95.8 | 94.2 | |
| π0Method Category=VLA-Based2026.06 | 85.2 | 96.8 | 98.8 | 95.8 | 94.2 | |
| π0Scale (B)=32026.06 | 85.2 | 96.8 | 98.8 | 95.8 | 94.2 | |
| π0Guidance=Linguistics2026.07 | 85.2 | 96.8 | 98.8 | 95.8 | 94.1 | |
| DivPruneSpeedup=1.46×2026.06 | 84.8 | 92.4 | 91.2 | 89 | 89.4 | |
| ADPSpeedup=1.35×2026.06 | 84.2 | 97.6 | 98.4 | 97.4 | 94.4 | |
| π0-SpecPrune-VLASpeedup=1.31×2026.06 | 84.2 | 96.6 | 98 | 95.2 | 93.5 | |
| π0-FAST + MG-Selectjoint training=true, test-time scaling=true2025.10 | 82.7 | 97.2 | 98 | 94.5 | 93.1 | |
| VLA-RLReward=Env. Success2026.06 | 82.2 | 90.2 | 94.3 | 91.8 | 89.6 | |
| FPC-VLAGuidance=Linguistics2026.07 | 82.2 | 87 | 92 | 86.2 | 86.9 | |
| GraspVLAArchitecture=Enhanced VLA2026.06 | 82 | — | 94.1 | 91.2 | 89.1 | |
| π0-FASTreproduced=true2025.10 | 79.6 | 97.4 | 95.4 | 95.6 | 92 | |
| 4D-VLAScale (B)=42026.06 | 79.1 | 88.9 | 95.2 | 90.9 | 88.6 | |
| SeerScale (B)=0.572026.06 | 78.7 | — | — | — | 78.7 | |
| MolmoActScale (B)=72026.06 | 77.2 | 87 | 95.4 | 87.6 | 86.6 | |
| SmolVLAArchitecture=Enhanced VLA2026.06 | 77 | 93 | 94 | 91 | 88.8 | |
| SmolVLAScale (B)=2.22026.06 | 77 | 93 | 94 | 91 | 88.8 |