Robot Manipulation on LIBERO
100Object AchievementCosmos Policy
Evaluation results
| Method | Links | |||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Cosmos Policy2026.01 | 100 | 98.1 | 98.2 | 97.6 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Xiaomi-Robotics-0Evaluation protocol=Standard, Action chunk length (T)=102026.02 | 100 | 98.8 | 98.8 | 97.2 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AcceRL2026.03 | 100 | 99.6 | 98.8 | 99.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BF16Precision=16-bit2026.03 | 100 | 97.5 | 97 | — | 97.75 | — | — | — | — | — | — | — | — | — | 96.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AWQCalibration dataset=Spatial, Bits (q)=2 bits, Groupsize (g)=642026.03 | 100 | 90.5 | 85 | — | 89.34 | — | — | — | — | — | — | — | — | — | 82 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AWQCalibration dataset=Goal, Bits (q)=2 bits, Groupsize (g)=642026.03 | 100 | 92.5 | 93.5 | — | 92.63 | — | — | — | — | — | — | — | — | — | 84.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| TTQRank (r)=16, Bits (q)=2 bits, Groupsize (g)=642026.03 | 100 | 94.5 | 93.5 | — | 93.88 | — | — | — | — | — | — | — | — | — | 87.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FocusVLAParam=0.5B, Policy Weighting Strategy=Multi-weights (one policy per task)2026.03 | 100 | 99.6 | 98.8 | 96.2 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MWMModality=Mask, Cameras=3rd+wrist2026.04 | 100 | 98.8 | 98.2 | — | 98.3 | — | — | — | — | — | — | — | — | — | 96 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cosmos PolicyAux. Mech.=Plan+WM, NFE=102026.04 | 100 | 98.1 | 98.2 | 97.6 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| πRLParadigm=RL, Warm-up Strategy=Single-trajectory, Camera Configuration=Two-camera2026.04 | 100 | 99.6 | 99.6 | 94 | 98.3 | 2 | 1 | 2 | 3 | 2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LaST-R1Paradigm=RL, Warm-up Strategy=Single-trajectory, Camera Configuration=Single-view2026.04 | 100 | 99.8 | 100 | 99.4 | 99.8 | 1 | 1 | 1 | 1 | 1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AttenA+OFT2026.05 | 100 | 99 | 98.8 | — | — | — | — | — | — | — | — | — | — | — | 96.6 | — | — | — | — | — | — | — | — | 98.6 | 1.4 | — | — | |
| 3DThinkVLA2026.06 | 100 | 100 | 98.8 | 95.8 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Fast-WAMActive Params=6B, Embodied Pretraining=false2026.06 | 100 | 98.2 | 97 | 95.2 | 97.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WLA-0Active Params=2B, Embodied Pretraining=false2026.06 | 100 | 99 | 97.8 | 97.6 | 98.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| +TTSActive Params=2B, Embodied Pretraining=false, test-time scaling=true2026.06 | 100 | 99.2 | 98.4 | 97.8 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FLOWER+IVRAinference-time modification=IVRA, token mixing coefficient lambda=0.22026.01 | 99.9 | 97.3 | 97.6 | 94.9 | 97.1 | — | — | — | — | — | — | — | — | — | — | — | 96 | — | — | — | — | — | — | — | — | — | — | |
| ABot-M0training=jointly on all suites2026.02 | 99.8 | 98.8 | 99 | 96.6 | 98.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EO12026.01 | 99.8 | 99.7 | 99.2 | 94.8 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ActionCodec-BARPt.=N2026.02 | 99.8 | 97.2 | 97.6 | 94.8 | 97.4 | — | — | — | — | 2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ActionCodec-BARPt.=N2026.02 | 99.8 | 97.2 | 97.6 | 94.8 | 97.4 | — | — | — | — | 1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VOTEMethod Category=Basic 2D VLA2026.02 | 99.8 | 98.8 | 97.6 | 95.6 | 98 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 3D-CAVLAMethod Category=3D VLA with explicit 3D inputs2026.02 | 99.8 | 98.2 | 98.2 | 96.1 | 98.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ROCKETMethod Category=3D VLA with representation alignment2026.02 | 99.8 | 98.2 | 98.8 | 97 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 3D-CAVLA2026.02 | 99.8 | 98.2 | 98.2 | 96.1 | 98.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SimVLAB (Backbone Scale in billions)=0.52026.02 | 99.8 | 99.6 | 98.6 | 96.4 | 98.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OptimusVLA2026.02 | 99.8 | 99.6 | 98.4 | 96.4 | 98.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EO-1Evaluation protocol=Standard, Action chunk length (T)=102026.02 | 99.8 | 99.7 | 99.2 | 94.8 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FutureVLA-GTArchitecture=GR00T-style2026.03 | 99.8 | 98.8 | 98.6 | 96 | 98.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SimpleVLA-RL2026.03 | 99.8 | 99.4 | 99.2 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RLinf-VLA2026.03 | 99.8 | 99.4 | 98.8 | 94 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| π0.5 + TAG-blackGuidance Mechanism=TAG-black2026.03 | 99.8 | 97.6 | 97.4 | 95.8 | 97.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| A12026.04 | 99.8 | 97.4 | 97.6 | 91.4 | 96.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EO-1Pretraining=Yes, Action Type=Continuous2025.11 | 99.8 | 99.7 | 99.2 | 94.8 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RynnVLA-002-ContinuousPretraining=No, Action Type=Continuous2025.11 | 99.8 | 99 | 96.4 | 94.4 | 97.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| A12026.06 | 99.8 | 97.4 | 97.6 | 91.4 | 96.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 3D-CAVLA2026.06 | 99.8 | 98.2 | 98.2 | 96.1 | 98.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MotusActive Params=8B, Embodied Pretraining=true2026.06 | 99.8 | 96.8 | 96.6 | 97.6 | 97.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DAG-VLABackbone=π₀.₅, Guidance=AFIL2026.05 | 99.7 | 99.8 | 98.1 | — | 98.4 | — | — | — | — | — | — | — | — | — | 95.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CronusVLA2026.01 | 99.6 | 97.3 | 96.9 | 94 | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LingBot-VA2026.01 | 99.6 | 98.5 | 97.2 | 98.5 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HoloBrain-0-QWSize=1.1B2026.02 | 99.6 | 97.2 | 97.6 | 95.2 | 97.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Being-H0.5variant=specialist2026.01 | 99.6 | 99.2 | 99.4 | 97.4 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ACoT-VLAGuidance=Action2026.01 | 99.6 | 99.4 | 98.8 | 96 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LoLA2025.12 | 99.6 | 99.6 | 97.2 | 88.2 | 96.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ActionCodec (2.2B)Pt.=N2026.02 | 99.6 | 96.2 | 95.4 | 90.6 | 95.5 | — | — | — | — | 5.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ActionCodec (2.2B)Pt.=N, Backbone=SmolVLM2-2.2B2026.02 | 99.6 | 96.2 | 95.4 | 90.6 | 95.5 | — | — | — | — | 3.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CronusVLAMethod Category=Basic 2D VLA2026.02 | 99.6 | 97.3 | 96.9 | 94 | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Spatial ForcingMethod Category=3D VLA with representation alignment2026.02 | 99.6 | 99.4 | 98.8 | 96 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AtomVLA (SFT + Post-Training)w/o Pre-Training=✗, Model Size=4 B2026.03 | 99.6 | 96.4 | 97.6 | 94.4 | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Spatial ForcingParam=7B, Policy Weighting Strategy=Multi-weights (one policy per task)2026.03 | 99.6 | 99.4 | 98.8 | 96 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VLA-Adapter-ProParam=0.5B, Policy Weighting Strategy=Multi-weights (one policy per task)2026.03 | 99.6 | 99.6 | 98.2 | 96.4 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Spatial ForcingParam=72026.04 | 99.6 | 99.4 | 98.8 | 96 | — | — | — | — | — | — | — | — | — | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VLA-AdapterParam=1.222026.04 | 99.6 | 99.6 | 98.2 | 96.4 | — | — | — | — | — | — | — | — | — | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PokeVLAParam=1.222026.04 | 99.6 | 99.6 | 98.4 | 95.2 | — | — | — | — | — | — | — | — | — | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InstructVLAAux. Mech.=-, NFE=102026.04 | 99.6 | 97.3 | 96.5 | 89.8 | 95.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CronusVLAAux. Mech.=7B, NFE=1002026.04 | 99.6 | 97.3 | 96.9 | 94 | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VLA-JEPAModel Category=World-Action Models2026.05 | 99.6 | 96.2 | 97.2 | 95.8 | 97.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OneWM-VLA2026.05 | 99.6 | 98.2 | 99 | 95.6 | 98.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| π0 + ALAM (ours)Type=LA, Size=3B2026.05 | 99.6 | 99.2 | 99 | 94.4 | 98.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RotVLASize=1.7B2026.05 | 99.6 | 98.2 | 98.4 | 96.4 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpatialForcing2026.06 | 99.6 | 99.4 | 98.8 | 96 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Lingbot-VAActive Params=5B, Embodied Pretraining=true2026.06 | 99.6 | 98.5 | 97.2 | 98.5 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| -LwmActive Params=2B, Embodied Pretraining=false, World Expert Loss=false2026.06 | 99.6 | 98.4 | 97 | 96.4 | 97.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AWQCalibration dataset=10, Bits (q)=2 bits, Groupsize (g)=642026.03 | 99.5 | 94 | 92.5 | — | 90.63 | — | — | — | — | — | — | — | — | — | 76.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| TTQRank (r)=0, Bits (q)=2 bits, Groupsize (g)=642026.03 | 99.5 | 93 | 93.5 | — | 93.25 | — | — | — | — | — | — | — | — | — | 87 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoLA (Ours)Fine-tuned=true2026.05 | 99.5 | 93 | — | 96 | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| StreamVLAScale=Medium, Params (B)=32026.02 | 99.4 | 99.2 | 98.6 | 96.6 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ΔVLA2026.03 | 99.4 | 98.6 | 97.4 | 95.6 | 97.8 | 1 | 1 | 2 | 1 | 1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| E02025.11 | 99.4 | 97.2 | 95 | 92.2 | 96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GR00T N1.5status=reproduced2025.10 | 99.4 | 98.2 | 97.2 | 87.8 | 95.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GR00T N1.52026.06 | 99.4 | 98.2 | 97.2 | — | 95.7 | — | — | — | — | — | — | — | — | — | 87.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FLOWER2026.01 | 99.3 | 97.2 | 96.9 | 94.5 | 96.3 | — | — | — | — | — | — | — | — | — | — | — | 93.4 | — | — | — | — | — | — | — | — | — | — | |
| Dream4manipParam=1.222026.02 | 99.2 | 99.4 | 98.2 | 95.4 | — | — | — | — | — | — | — | — | — | 98.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| π0.5strategy=joint learning2025.12 | 99.2 | 98.4 | 96 | 96.2 | 97.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VLA-AdapterB (Backbone Scale in billions)=0.52026.02 | 99.2 | 97.8 | 97.2 | 95 | 97.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FutureVLA-OTArchitecture=OFT-style2026.03 | 99.2 | 99.6 | 99.2 | 94.8 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| π0.5 + TAG-bgGuidance Mechanism=TAG-bg2026.03 | 99.2 | 97.8 | 97.4 | 97 | 97.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cosmos w/ Latent IDMModality=RGB, Cameras=3rd2026.04 | 99.2 | 94.8 | 89.2 | — | 91.9 | — | — | — | — | — | — | — | — | — | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| VLA-Adapter2026.04 | 99.2 | 97.8 | 97.2 | 95 | 97.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CF-VLAAux. Mech.=3B, NFE=22026.04 | 99.2 | 98 | 96.6 | 92 | 96.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FIBERTUNEStatus=pi0.5 adapted polish2026.06 | 99.2 | 97.6 | 97.8 | — | 97.65 | — | — | — | — | — | — | — | — | — | 96 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FLOWER2026.02 | 99.1 | 97.5 | 96.1 | 94.9 | 96.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FLOWEREvaluation protocol=Standard, Action chunk length (T)=102026.02 | 99.1 | 97.5 | 96.1 | 94.9 | 96.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Flower VLAAux. Mech.=-, NFE=82026.04 | 99.1 | 97.5 | 96.1 | 94.9 | 96.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternVLA-M12026.02 | 99 | 98 | 93.8 | 92.6 | 95.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LAP-3BRank=52026.02 | 99 | 98.2 | 98.8 | — | 96.8 | — | — | — | — | — | — | — | — | — | 91.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LAP-3B+ VQA Co.VQA Co-training=true, Rank=22026.02 | 99 | 99 | 97.2 | — | 97.2 | — | — | — | — | — | — | — | — | — | 93.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternVLA-M12026.01 | 99 | 98 | 93.8 | 92.6 | 95.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternVLAMethod Category=Basic 2D VLA2026.02 | 99 | 98 | 93.8 | 92.6 | 95.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GeoVLAMethod Category=3D VLA with explicit 3D inputs2026.02 | 99 | 98.4 | 96.6 | 96.6 | 97.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| π0.5Precision=FP162026.02 | 99 | 98.5 | 97.5 | 93.5 | 97.1 | — | — | — | — | — | — | 4.27 | — | — | — | — | — | — | 0 | — | — | — | — | — | — | — | — | |
| QuantVLA (LLM)Base Model=π0.5, Precision=W4A82026.02 | 99 | 98.5 | 96.5 | 96.5 | 97.6 | — | — | — | — | — | — | 1.58 | — | — | — | — | — | — | 63 | — | — | — | — | — | — | — | — | |
| VLA-Thinker2026.03 | 99 | 98.7 | 95.2 | 96.9 | 97.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternVLA-M12026.03 | 99 | 98 | 93.8 | 92.6 | 95.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepThinkVLACategory=Hybrid-Decoding, Variant=π0-FAST, Input Modality=vision-language only, Trials per task=502025.10 | 99 | 96.6 | 96.4 | 96.2 | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternVLA-M1Model Category=Vision-Language-Action Models2026.05 | 99 | 98 | 93.8 | 92.6 | 95.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OA-WAMModel Category=World-Action Models2026.05 | 99 | 98.9 | 97.4 | 95.9 | 97.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GeoVLA2026.06 | 99 | 98.4 | 96.6 | 96.6 | 97.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |