Robot Manipulation on MetaWorld (Score)
87.8ScoreToBo
Evaluation results
| Method | Links | |
|---|---|---|
| ToBo#Param=21.7M, Pre-training Dataset=Kinetics-400, #Seen frames=0.2B, Supervision Type=Self-supervised Learning2025.07 | 87.8 | |
| data4robotics#Param=86.0M, Pre-training Dataset=Kinetics-700, #Seen frames=0.5B, Supervision Type=Self-supervised Learning2025.07 | 87 | |
| VC-1#Param=86.0M, Pre-training Dataset=Ego4D+N, #Seen frames=1.0B, Supervision Type=Self-supervised Learning2025.07 | 86.4 | |
| Theia#Param=52.9M, Pre-training Dataset=Theia dataset, #Seen frames=14.4B*, Supervision Type=Supervision through Foundation Models, Uses additional compression layers=true2025.07 | 86.1 | |
| MPI#Param=21.7M, Pre-training Dataset=Ego4D, #Seen frames=0.1B, Supervision Type=Supervision with Auxiliary Language Guidance, Uses multi-head attention pooling layers=true2025.07 | 85.7 | |
| MVP#Param=21.7M, Pre-training Dataset=MVP dataset, #Seen frames=4.8B, Supervision Type=Supervision with Auxiliary Language Guidance2025.07 | 84.6 | |
| R3M#Param=25.6M, Pre-training Dataset=Ego4D, #Seen frames=0.8B, Supervision Type=Supervision with Auxiliary Language Guidance2025.07 | 69.2 | |
| Voltron#Param=21.7M, Pre-training Dataset=SS-v2, #Seen frames=0.3B, Supervision Type=Supervision with Auxiliary Language Guidance, Uses multi-head attention pooling layers=true2025.07 | 68.7 | |
| R3M#Param=25.6M, Pre-training Dataset=Ego4D, #Seen frames=0.8B, Supervision Type=Self-supervised Learning, Excludes language guidance=true2025.07 | 67 |