Robot Manipulation on MetaWorld, Adroit, and Dexart Combined
87.2Average Success RateX-Distill
Evaluation results
| Method | Links | |
|---|---|---|
| X-Distill2026.01 | 87.2 | |
| PointNet-DP3input=privileged background-cropped 3D point clouds2026.01 | 84 | |
| IDPNFE=1, Hardware=NVIDIA A402026.05 | 81.3 | |
| FlowNFE=10, Hardware=NVIDIA A402026.05 | 79.9 | |
| Naive DriftingNFE=1, Hardware=NVIDIA A402026.05 | 79.8 | |
| MP1NFE=1, Hardware=NVIDIA A402026.05 | 79.7 | |
| DP3NFE=10, Hardware=NVIDIA A402026.05 | 79.4 | |
| FlowPolicyNFE=1, Hardware=NVIDIA A402026.05 | 76.9 | |
| DP3NFE=10, Source=Cited from DP3 [33]2026.05 | 73.9 | |
| OFPNFE=1, Source=Cited from OFP [20]2026.05 | 71.6 | |
| CPNFE=1, Hardware=NVIDIA A402026.05 | 68.4 | |
| DINOv22026.01 | 66.2 | |
| ResNet-scratch2026.01 | 64.1 | |
| X-DP3Horizon (H)=16, Conditional Trajectory Length (N)=8, Number of trials=300, Checkpoint strategy=Final checkpoint2026.05 | 63.8 | |
| OneDPNFE=1, Source=Cited from OFP [20]2026.05 | 62.4 | |
| DP3+HybridHorizon (H)=16, Conditional Trajectory Length (N)=8, Number of trials=300, Checkpoint strategy=Final checkpoint2026.05 | 60.4 | |
| DP3+SepAuxHorizon (H)=16, Conditional Trajectory Length (N)=8, Number of trials=300, Checkpoint strategy=Final checkpoint2026.05 | 60.2 | |
| DP3+MBAHorizon (H)=16, Conditional Trajectory Length (N)=8, Number of trials=300, Checkpoint strategy=Final checkpoint2026.05 | 58.9 | |
| FlowNFE=10, Source=Cited from OFP [20]2026.05 | 57.9 | |
| MP1NFE=1, Source=Cited from OFP [20]2026.05 | 57.4 | |
| DP3+UniAuxHorizon (H)=16, Conditional Trajectory Length (N)=8, Number of trials=300, Checkpoint strategy=Final checkpoint2026.05 | 56.3 | |
| Depth-Anything2026.01 | 56.1 | |
| CPNFE=1, Source=Cited from OFP [20]2026.05 | 54.7 | |
| DP3Horizon (H)=16, Conditional Trajectory Length (N)=8, Number of trials=300, Checkpoint strategy=Final checkpoint2026.05 | 54.6 | |
| Theia2026.01 | 36 |