Robotic Manipulation on RLBench
96Place Cups SuccessMALLVi
Evaluation results
| Method | Links | ||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MALLViReflector Agent=true2026.02 | 96 | — | — | — | — | — | — | — | — | — | — | 94 | 92 | — | — | — | 90 | 83 | — | — | — | — | |
| MALLViReflector Agent=false2026.02 | 75 | — | — | — | — | — | — | — | — | — | — | 89 | 81 | — | — | — | 78 | 63 | — | — | — | — | |
| ActiveVLA2026.01 | 65.6 | 0.918 | 1.22 | 100 | 100 | 92.4 | 100 | 100 | 96.2 | 100 | 87.4 | 99.5 | 99.4 | 89.3 | 98.5 | 63.3 | 82.2 | 84.8 | 100 | 95.3 | — | — | |
| BridgeVLA2026.01 | 58.4 | 0.882 | 2.44 | 100 | 100 | 88 | 100 | 100 | 88 | 98.4 | 73.6 | 99.2 | 99.2 | 87.2 | 96 | 60.8 | 76.8 | 81.6 | 87.2 | 92.8 | — | — | |
| HDFlow2026.05 | 56.7 | — | — | 96 | 100 | 93.3 | 98.7 | 82 | 94.7 | 100 | 72 | 98.7 | 96.7 | 88.7 | 91.3 | 73.3 | 56 | 81.3 | 100 | 95.3 | — | — | |
| ARP+2025.01 | 48.8 | 84.9 | 3.2 | 95.2 | 99.2 | 78.4 | 97.6 | 92.8 | 96 | 100 | 69.6 | 98.4 | 86.4 | 89.6 | 92.8 | 46.4 | 63.2 | 80 | 97.6 | 96 | — | — | |
| SAM2Act2025.01 | 47 | 86.8 | 3.1 | 99 | 99 | 84 | 98 | 83 | 93 | 100 | 75 | 99 | 98 | 89 | 86 | 64 | 76 | 78 | 99 | 96 | — | — | |
| 3D-LOTUS2025.01 | 40.8 | 83.1 | 3.7 | 96 | 100 | 69.6 | 98.4 | 85.6 | 91.2 | 100 | 78.4 | 97.6 | 95.2 | 88.8 | 99.2 | 34.4 | 58.4 | 75.2 | 96 | 90.4 | — | — | |
| RVT-22025.01 | 38 | 81.4 | 3.7 | 100 | 99 | 40 | 99 | 74 | 95 | 100 | 66 | 96 | 96 | 88 | 92 | 35 | 80 | 69 | 100 | 99 | — | — | |
| RVT-22026.01 | 38 | 0.814 | 3 | 100 | 99 | 40 | 99 | 74 | 95 | 100 | 66 | 96 | 96 | 88 | 92 | 35 | 80 | 69 | 100 | 99 | — | — | |
| RVT-22026.05 | 38 | — | — | 100 | 99 | 40 | 99 | 74 | 95 | 100 | 66 | 96 | 96 | 88 | 92 | 35 | 80 | 69 | 100 | 99 | — | — | |
| 3D Diffuser Actor2025.01 | 24 | 81.3 | 3.9 | 96 | 100 | 65.6 | 96.8 | 89.6 | 93.6 | 98.4 | 85.6 | 96 | 97.6 | 82.4 | 97.6 | 44 | 68.3 | 47.2 | 84 | 99.2 | — | — | |
| 3D Diffuser Actor2026.01 | 24 | 0.813 | 3.39 | 96 | 100 | 65.6 | 96.8 | 89.6 | 93.6 | 98.4 | 85.6 | 96 | 97.6 | 82.4 | 97.6 | 44 | 68.3 | 47.2 | 84 | 99.2 | — | — | |
| 3D Actor2026.05 | 24 | — | — | 96 | 100 | 65.6 | 96.8 | 89.6 | 93.6 | 98.4 | 85.6 | 96 | 97.6 | 82.4 | 97.6 | 44 | 68.3 | 47.2 | 84 | 99.2 | — | — | |
| Single-AgentMulti-Agent Decomposition=false2026.02 | 22 | — | — | — | — | — | — | — | — | — | — | 73 | 58 | — | — | — | 42 | 15 | — | — | — | — | |
| SAM2ActPretrain [Type]=Supervised, Finetune [Method]=RVT2026.03 | 9 | 68 | — | 71 | 100 | 13 | 95 | 79 | 93 | 100 | 55 | 94 | 89 | 58 | 66 | 25 | 46 | 63 | 80 | 88 | — | — | |
| HyperMVPPretrain [Type]=Self-Supervised, Finetune [Method]=RVT2026.03 | 7 | 71.1 | — | 80 | 98 | 12 | 97 | 75 | 94 | 100 | 59 | 93 | 97 | 61 | 90 | 26 | 43 | 64 | 88 | 96 | — | — | |
| RVT2025.01 | 4 | 62.9 | 6.9 | 52 | 99.2 | 11.2 | 88 | 71.2 | 91 | 100 | 49.6 | 88 | 91.2 | 48 | 81.6 | 36 | 28.8 | 26.4 | 72 | 93.6 | — | — | |
| RVTTrain time (in days)=1, Inf. Speed (in fps)=11.62023.06 | 4 | 62.9 | 1.1 | 52 | 99.2 | 11.2 | 88 | 71.2 | 91 | 100 | 49.6 | 88 | 91.2 | 48 | 81.6 | 36 | 28.8 | 26.4 | 72 | 93.6 | — | — | |
| RVTInference Steps (Sum)=6158±642024.05 | 4 | 62.9 | — | 52 | — | 11.2 | 88 | 71.2 | 91 | 100 | 49.6 | 89.6 | 91.2 | 48 | — | 36 | 28.8 | 26.4 | 72 | 93.6 | 99.2 | 81.6 | |
| RVT2024.06 | 4 | 62.9 | — | 52 | 99.2 | 11.2 | 88 | 71.2 | 91 | 100 | 49.6 | 88 | 91.2 | 48 | 81.6 | 36 | 28.8 | 26.4 | 72 | 93.6 | — | — | |
| 3D-MVP2024.06 | 4 | 67.5 | — | 76 | 100 | 20 | 96 | 84 | 100 | 96 | 60 | 100 | 92 | 60 | 48 | 28 | 40 | 36 | 80 | 96 | — | — | |
| Image-BCBackbone=CNN2026.01 | 4 | 0.013 | 11.56 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 8 | 4 | 0 | 0 | 0 | 0 | 0 | 0 | 8 | — | — | |
| RVT2026.01 | 4 | 0.629 | 5.39 | 52 | 99.2 | 11.2 | 88 | 71.2 | 91 | 100 | 49.6 | 88 | 91.2 | 48 | 81.6 | 36 | 28.8 | 26.4 | 72 | 93.6 | — | — | |
| RVTPretrain [Type]=-, Finetune [Method]=-2026.03 | 4 | 62.9 | — | 52 | 99.2 | 11.2 | 88 | 71.2 | 91 | 100 | 49.6 | 88 | 91.2 | 48 | 81.6 | 36 | 28.8 | 26.4 | 72 | 93.6 | — | — | |
| 3D-MVPPretrain [Type]=Self-Supervised, Finetune [Method]=RVT2026.03 | 4 | 67.5 | — | 76 | 100 | 20 | 96 | 84 | 100 | 96 | 60 | 100 | 92 | 60 | 48 | 28 | 40 | 36 | 80 | 96 | — | — | |
| RVT2026.05 | 4 | — | — | 52 | 99.2 | 11.2 | 88 | 71.2 | 91 | 100 | 49.6 | 88 | 91.2 | 48 | 81.6 | 36 | 28.8 | 26.4 | 72 | 93.6 | — | — | |
| Act3D2025.01 | 3 | 65 | 6.8 | 92 | 92 | 27 | 94 | 93 | 80 | 99 | 51 | 90 | 95 | 47 | 93 | 8 | 12 | 9 | 92 | 94 | — | — | |
| Act3D2026.01 | 3 | 0.65 | 5.28 | 92 | 92 | 27 | 94 | 93 | 80 | 99 | 51 | 90 | 95 | 47 | 93 | 8 | 12 | 9 | 92 | 94 | — | — | |
| PerAct2025.01 | 2.4 | 49.4 | 8.9 | 55.2 | 89.6 | 5.6 | 70.4 | 88 | 44.8 | 92.8 | 28 | 51.2 | 84 | 17.6 | 74 | 16.8 | 26.4 | 2.4 | 52 | 88 | — | — | |
| PerActTrain time (in days)=16, Inf. Speed (in fps)=4.92023.06 | 2.4 | 49.4 | 1.9 | 55.2 | 89.6 | 5.6 | 70.4 | 88 | 44.8 | 92.8 | 28 | 51.2 | 84 | 17.6 | 74 | 16.8 | 26.4 | 2.4 | 52 | 88 | — | — | |
| PerAct2024.05 | 2.4 | 49.4 | — | 55.2 | — | 5.6 | 70.4 | 88 | 44.8 | 92.8 | 28 | 51.2 | 84 | 17.6 | — | 16.8 | 26.4 | 2.4 | 52 | 88 | 59.6 | 74 | |
| PerAct2024.06 | 2.4 | 49.4 | — | 55.2 | 89.6 | 5.6 | 70.4 | 88 | 44.8 | 92.8 | 28 | 51.2 | 84 | 17.6 | 74 | 16.8 | 26.4 | 2.4 | 52 | 88 | — | — | |
| PerAct2026.01 | 2.4 | 0.494 | 7.33 | 55.2 | 89.6 | 5.6 | 70.4 | 88 | 44.8 | 92.8 | 28 | 51.2 | 84 | 17.6 | 74 | 16.8 | 26.4 | 2.4 | 52 | 88 | — | — | |
| PerActPretrain [Type]=-, Finetune [Method]=-2026.03 | 2.4 | 49.4 | — | 55.2 | 89.6 | 5.6 | 70.4 | 88 | 44.8 | 92.8 | 28 | 51.2 | 84 | 17.6 | 74 | 16.8 | 26.4 | 2.4 | 52 | 88 | — | — | |
| PerAct2026.05 | 2.4 | — | — | 55.2 | 89.6 | 5.6 | 70.4 | 88 | 44.8 | 92.8 | 28 | 51.2 | 84 | 17.6 | 74 | 16.8 | 26.4 | 2.4 | 52 | 88 | — | — | |
| SAM-E (SAM -> R3M)Backbone=SAM -> R3M, Inference Steps (Sum)=1165±632024.05 | 1.6 | 66.5 | — | 78.4 | — | 56 | 97.6 | 95.2 | 60 | 96 | 41.6 | 88.8 | 95.2 | 96.8 | — | 31.2 | 32 | 2.4 | 5.6 | 100 | 100 | 88.8 | |
| SAM-E (SAM -> DINO)Backbone=SAM -> DINO, Inference Steps (Sum)=1143±152024.05 | 0.8 | 67.1 | — | 89.6 | — | 11.2 | 96 | 95.2 | 93.6 | 100 | 52 | 88 | 85.6 | 64 | — | 30.4 | 28.8 | 1.6 | 100 | 100 | 100 | 89.6 | |
| Image-BCBackbone=CNN2025.01 | 0 | 1.3 | 12.4 | 0 | 0 | 0 | 0 | 4 | 0 | 0 | 0 | 8 | 4 | 0 | 0 | 0 | 0 | 0 | 0 | 8 | — | — | |
| Image-BCBackbone=ViT2025.01 | 0 | 1.3 | 12.6 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 16 | — | — | |
| C2F-ARM-BC2025.01 | 0 | 20.1 | 11.5 | 24 | 24 | 4 | 20 | 20 | 8 | 72 | 0 | 4 | 12 | 8 | 16 | 8 | 0 | 0 | 0 | 68 | — | — | |
| HiveFormer2025.01 | 0 | 45.3 | 9.6 | 52 | 76 | 0 | 100 | 52 | 80 | 84 | 32 | 68 | 76 | 8 | 64 | 8 | 8 | 0 | 28 | 80 | — | — | |
| PolarNet2025.01 | 0 | 46.4 | 9.1 | 36 | 92 | 4 | 100 | 84 | 40 | 96 | 12 | 32 | 84 | 44 | 56 | 12 | 4 | 8 | 52 | 80 | — | — | |
| SAM-E2025.01 | 0 | 70.6 | 4.8 | 82.4 | 100 | 18.4 | 95.2 | 95.2 | 94.4 | 100 | 64 | 92 | 95.2 | 78.4 | 95.2 | 34.4 | 26.4 | 0 | 100 | 100 | — | — | |
| Image-BCBackbone=CNN2023.06 | 0 | 1.3 | 3.7 | 0 | 0 | 0 | 0 | 4 | 0 | 0 | 0 | 8 | 4 | 0 | 0 | 0 | 0 | 0 | 0 | 8 | — | — | |
| Image-BCBackbone=ViT2023.06 | 0 | 1.3 | 3.8 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 16 | — | — | |
| C2F-ARM-BC2023.06 | 0 | 20.1 | 3.1 | 24 | 24 | 4 | 20 | 20 | 8 | 72 | 0 | 4 | 12 | 8 | 16 | 8 | 0 | 0 | 0 | 68 | — | — | |
| SAM-E (SAM -> RVT)Backbone=SAM -> RVT, Inference Steps (Sum)=1190±192024.05 | 0 | 65.3 | — | 83.2 | — | 7.2 | 95.2 | 95.2 | 92.8 | 98.4 | 59.2 | 88 | 72 | 61.6 | — | 35.2 | 22.4 | 3.2 | 84 | 87.2 | 99.2 | 79.2 | |
| SAM-E (SAM -> CLIP)Backbone=SAM -> CLIP, Inference Steps (Sum)=1192±172024.05 | 0 | 64.8 | — | 73.6 | — | 8 | 96 | 92 | 93.6 | 100 | 40 | 78.4 | 80.8 | 59.2 | — | 42.4 | 22.4 | 4.8 | 85.6 | 100 | 99.2 | 78.4 | |
| SAM-E (ours)Inference Steps (Sum)=1130±122024.05 | 0 | 70.6 | — | 82.4 | — | 18.4 | 95.2 | 95.2 | 94.4 | 100 | 64 | 92 | 95.2 | 78.4 | — | 34.4 | 26.4 | 0 | 100 | 95.2 | 100 | 95.2 | |
| Image-BCBackbone=CNN2024.06 | 0 | 1.3 | — | 0 | 0 | 0 | 0 | 4 | 0 | 0 | 0 | 8 | 4 | 0 | 0 | 0 | 0 | 0 | 0 | 8 | — | — | |
| Image-BCBackbone=ViT2024.06 | 0 | 1.3 | — | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 16 | — | — | |
| C2F-ARM-BC2024.06 | 0 | 20.1 | — | 24 | 24 | 4 | 20 | 20 | 8 | 72 | 0 | 4 | 12 | 8 | 16 | 8 | 0 | 0 | 0 | 68 | — | — | |
| D4Radaptation=None2024.06 | 0 | 55.3 | — | 61.3 | 100 | 32 | 97.3 | 84 | 81.3 | 93.3 | 46.7 | 25.3 | 88 | 32 | 57.3 | 24 | 6.7 | 0 | 68 | 97.3 | — | — | |
| D4R-Alignadaptation=HR-Align2024.06 | 0 | 59.9 | — | 72 | 97.3 | 24 | 92 | 81.3 | 84 | 100 | 46.7 | 32 | 97.3 | 57.3 | 70.7 | 22.7 | 16 | 1.3 | 86.7 | 96 | — | — | |
| R3Madaptation=None2024.06 | 0 | 50.3 | — | 74.7 | 96 | 17.3 | 96 | 80 | 77.3 | 100 | 24 | 14.7 | 90.7 | 30.7 | 60 | 20 | 16 | 0 | 36 | 72 | — | — | |
| R3M-Alignadaptation=HR-Align2024.06 | 0 | 59.2 | — | 80 | 97.3 | 16 | 93.3 | 80 | 88 | 100 | 33.3 | 70.7 | 78.7 | 41.3 | 80 | 13.3 | 13.3 | 1.3 | 98.7 | 80 | — | — | |
| Image-BCBackbone=ViT2026.01 | 0 | 0.013 | 11.61 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 16 | — | — | |
| C2F-ARM-BC2026.01 | 0 | 0.201 | 9.89 | 24 | 24 | 4 | 20 | 20 | 8 | 72 | 0 | 4 | 12 | 8 | 16 | 8 | 0 | 0 | 0 | 68 | — | — | |
| HiveFormer2026.01 | 0 | 0.453 | 8.22 | 52 | 76 | 0 | 100 | 52 | 80 | 84 | 32 | 68 | 76 | 8 | 64 | 8 | 8 | 0 | 28 | 80 | — | — | |
| PolarNet2026.01 | 0 | 0.464 | 7.78 | 36 | 92 | 4 | 100 | 84 | 40 | 96 | 12 | 32 | 84 | 44 | 56 | 12 | 4 | 8 | 52 | 80 | — | — | |
| PerAct2026.02 | 0 | — | — | — | — | — | — | — | — | — | — | 68 | 44 | — | — | — | 36 | 0 | — | — | — | — | |
| PolarNetPretrain [Type]=-, Finetune [Method]=-2026.03 | 0 | 46.4 | — | 36 | 92 | 4 | 100 | 84 | 40 | 96 | 12 | 32 | 84 | 44 | 56 | 12 | 4 | 8 | 52 | 80 | — | — | |
| 4DGentype=Baseline2026.02 | — | 47 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Act-Headtype=Ablation, description=skips latent optimization and uses action-head prediction2026.02 | — | 72.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Full IDMtype=Ablation, description=point-based inverse dynamics model2026.02 | — | 68.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MVISTA-4Dtype=Full model2026.02 | — | 72.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| P-ACTtype=Baseline2026.02 | — | 60.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TesserActtype=Baseline2026.02 | — | 67.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniPi*type=Baseline2026.02 | — | 34.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| w/o R-IDMtype=Ablation, description=removes the residual IDM2026.02 | — | 69 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |