Robotic Manipulation on LIBERO-Goal
98.8Success RatePolicyTrim
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| PolicyTrimModel=π0.52026.06 | 98.8 | — | — | — | — | 5.23 | — | 63.5 | 15 | — | |
| BaselineModel=π0.52026.06 | 98.7 | — | — | — | — | 1 | — | 110.6 | 5 | — | |
| PolicyTrimModel=OpenVLA-OFT2026.06 | 98 | — | — | — | — | 1.77 | — | 66.9 | 8 | — | |
| BaselineModel=OpenVLA-OFT2026.06 | 97.7 | — | — | — | — | 1 | — | 118.6 | 8 | — | |
| BASEChunk Size=82026.04 | 96.8 | — | — | — | — | 1 | — | — | — | — | |
| GAM2026.05 | 96.3 | — | — | — | — | — | 1 | — | — | — | |
| π02026.05 | 95.8 | — | — | — | — | — | 2 | — | — | — | |
| CLIP-RT+AAFine-tuning protocol=Atomic Action Fine-tuning2025.12 | 95.3 | — | — | — | — | — | — | — | — | — | |
| MIPt* value=0.9, Finetuning steps=50k, Pre-trained policy=pi02025.12 | 95.2 | — | — | — | — | — | — | — | — | — | |
| FlowIntegration steps=10, Finetuning steps=50k, Pre-trained policy=pi02025.12 | 95 | — | — | — | — | — | — | — | — | — | |
| RCPLoss function=l2 loss, Finetuning steps=50k, Pre-trained policy=pi02025.12 | 94.6 | — | — | — | — | — | — | — | — | — | |
| SV-VLAChunk Size=642026.04 | 94.4 | — | — | — | — | 1.56 | — | — | — | — | |
| CLIP-RT+Fine-tuning protocol=Baseline2025.12 | 94.2 | — | — | — | — | — | — | — | — | — | |
| BASEChunk Size=642026.04 | 93.2 | — | — | — | — | 2.78 | — | — | — | — | |
| BEAST2026.05 | 93.1 | — | — | — | — | — | 3 | — | — | — | |
| π0-FAST2025.10 | 88.6 | — | — | — | — | — | — | — | — | — | |
| π0-FAST2026.05 | 88.6 | — | — | — | — | — | 4 | — | — | — | |
| RCPLoss function=l1 loss, Finetuning steps=50k, Pre-trained policy=pi02025.12 | 88 | — | — | — | — | — | — | — | — | — | |
| PolicyTrimModel=GR00T2026.06 | 86.3 | — | — | — | — | 2.08 | — | 60.8 | 10 | — | |
| PDFPub.=-, Param.=9M2026.04 | 86 | — | — | — | — | — | 1 | — | — | — | |
| OpenSPMParadigm=Flow matching, Parameters=0.24M, Block Length=52026.06 | 85.6 | — | — | — | — | — | — | — | — | 1,033.3 | |
| Ours2025.10 | 85.2 | — | — | — | — | — | 1 | — | — | — | |
| OpenVLAPub.=CoRL’24, Param.=-2026.04 | 85 | — | — | — | — | — | 2 | — | — | — | |
| OCTOPub.=RSS’24, Param.=93M2026.04 | 85 | — | — | — | — | — | 2 | — | — | — | |
| OctoTraining Protocol=SFT2025.10 | 84.6 | — | — | — | — | — | 2 | — | — | — | |
| Octo2026.05 | 84.6 | — | — | — | — | — | 5 | — | — | — | |
| Octo-BaseParadigm=Diffusion model, Parameters=~200M, Block Length=42026.06 | 84.6 | — | — | — | — | — | — | — | — | 54.7 | |
| BaselineModel=GR00T2026.06 | 84.2 | — | — | — | — | 1 | — | 63.3 | 5 | — | |
| WorldVLAParadigm=WAM, Parameters=7B, Block Length=252026.06 | 83.4 | — | — | — | — | — | — | — | — | 2.4 | |
| GRAPETraining Protocol=DPO2025.10 | 82.2 | — | — | — | — | — | 3 | — | — | — | |
| VLA-RL2025.10 | 82.2 | — | — | — | — | — | 3 | — | — | — | |
| MTLPub.=NeurIPS’23, Param.=-2026.04 | 80 | — | — | — | — | — | 3 | — | — | — | |
| FP ModelType=Stat., Prec.=BF16, Spd.=1.00×, Mem. (GB)=15.22026.03 | 79.2 | — | — | — | — | — | — | — | — | — | |
| OpenVLATraining Protocol=SFT2025.10 | 79.2 | — | — | — | — | — | 5 | — | — | — | |
| OpenVLA2026.05 | 79.2 | — | — | — | — | — | 6 | — | — | — | |
| OpenVLAParadigm=VLA, Parameters=7B, Block Length=12026.06 | 79.2 | — | — | — | — | — | — | — | — | 4.9 | |
| OpenVLA-DPOPub.=Arxiv’25, Param.=130M2026.04 | 79 | — | — | — | — | — | 4 | — | — | — | |
| QVLAType=Stat., Prec.=W4A4, Spd.=1.49×, Mem. (GB)=4.32026.03 | 78.8 | — | — | — | — | — | — | — | — | — | |
| SpatialVLA2026.05 | 78.6 | — | — | — | — | — | 7 | — | — | — | |
| SpatialVLAParadigm=VLA, Parameters=4B, Block Length=42026.06 | 78.6 | — | — | — | — | — | — | — | — | 6.7 | |
| DyQ-VLAType=Dyn., Prec.=W4AX, Spd.=1.48×, Mem. (GB)=4.72026.03 | 78.5 | — | — | — | — | — | — | — | — | — | |
| ATMPub.=RSS’23, Param.=-2026.04 | 78 | — | — | — | — | — | 5 | — | — | — | |
| SFT-4LIBEROPub.=Arxiv’25, Param.=130M2026.04 | 77 | — | — | — | — | — | 5 | — | — | — | |
| OpenVLAPub.=CoRL’24, Param.=-, Reproduced status (†)=true2026.04 | 76 | — | — | — | — | — | 6 | — | — | — | |
| TraceVLAParadigm=VLA, Parameters=7B, Block Length=12026.06 | 75.1 | — | — | — | — | — | — | — | — | 2.2 | |
| PackNetPub.=CVPR’18, Param.=-2026.04 | 75 | — | — | — | — | — | 8 | — | — | — | |
| TraceVLAPub.=ICLR’25, Param.=130M2026.04 | 75 | — | — | — | — | — | 7 | — | — | — | |
| Speculative DecodingChunk Size=42026.04 | 74.4 | — | — | — | — | 1.42 | — | — | — | — | |
| MG-SelectPub.=Arxiv’25, Param.=130M2026.04 | 73 | — | — | — | — | — | 9 | — | — | — | |
| SmoothQuantType=Stat., Prec.=W4A4, Spd.=1.54×, Mem. (GB)=4.72026.03 | 69.6 | — | — | — | — | — | — | — | — | — | |
| Diffusion Policy2025.10 | 68.3 | — | — | — | — | — | 6 | — | — | — | |
| Diffusion PolicyParadigm=Diffusion model, Parameters=~260M, Block Length=82026.06 | 68.3 | — | — | — | — | — | — | — | — | 14.2 | |
| DPPub.=IJRR’25, Param.=-2026.04 | 68 | — | — | — | — | — | 10 | — | — | — | |
| ERPub.=Arxiv’19, Param.=-2026.04 | 49 | — | — | — | — | — | 11 | — | — | — | |
| SeqLPub.=NeurIPS’23, Param.=-2026.04 | 22 | — | — | — | — | — | 12 | — | — | — | |
| CLaD2026.03 | — | — | — | — | 94.3 | — | — | — | — | — | |
| GeoAwareVLAPerturbation Type=Agent Camera2026.03 | — | 88.4 | 86.2 | 85.2 | 86.6 | — | — | — | — | — | |
| OpenVLA2026.03 | — | — | — | — | 97.6 | — | — | — | — | — | |
| OpenVLA-OFTPerturbation Type=Agent Camera2026.03 | — | 84.4 | 75.8 | 68 | 76.1 | — | — | — | — | — | |
| Ours-OVPerturbation Type=Agent Camera2026.03 | — | 96.2 | 94 | 96.4 | 95.5 | — | — | — | — | — | |
| Ours-πPerturbation Type=Agent Camera2026.03 | — | 97.4 | 96.6 | 88.4 | 94.1 | — | — | — | — | — | |
| pi_02026.03 | — | — | — | — | 91.6 | — | — | — | — | — | |
| pi_0.52026.03 | — | — | — | — | 97.2 | — | — | — | — | — | |
| UniVLA2026.03 | — | — | — | — | 95.6 | — | — | — | — | — | |
| π*0.5Perturbation Type=Agent Camera2026.03 | — | 86.4 | 86 | 85 | 85.8 | — | — | — | — | — | |
| π0.5Perturbation Type=Agent Camera2026.03 | — | 88.8 | 66.8 | 48.4 | 68 | — | — | — | — | — |