Robot Policy Inference Efficiency on NVIDIA RTX 2080 (inference)
2.6Inference Time (ms)DMPO
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DMPOVision=light ViT, Action=MLP, Params=1.78M, Size=~28MB, Steps=12026.01 | 2.6 | 385 | 772 | |
| DMPOVision=light ViT, Action=MLP, Params=1.78M, Size=~28MB, Steps=22026.01 | 4.9 | 204 | 410 | |
| ShortCutVision=light ViT, Action=MLP, Params=1.78M, Size=~28MB, Steps=52026.01 | 12 | 83 | 167 | |
| DMPOVision=light ViT, Action=MLP, Params=1.78M, Size=~28MB, Steps=52026.01 | 12 | 83 | 167 | |
| MP1Vision=PointNet, Action=UNet, Params=255.80M, Size=~4GB, Steps=12026.01 | 21.4 | 47 | 94 | |
| CPVision=ResNet-18×2, Action=UNet, Params=284.86M, Size=~4.4GB, Steps=12026.01 | 35.5 | 28 | 56 | |
| MP1Vision=PointNet, Action=UNet, Params=255.80M, Size=~4GB, Steps=22026.01 | 41.6 | 24 | 48 | |
| ReFlowVision=light ViT, Action=MLP, Params=1.78M, Size=~28MB, Steps=202026.01 | 47.7 | 21 | 42 | |
| MP1Vision=PointNet, Action=UNet, Params=255.80M, Size=~4GB, Steps=52026.01 | 102.1 | 10 | 20 | |
| DPVision=ResNet-18×2, Action=UNet, Params=281.19M, Size=~4.4GB, Steps=10 (DDIM)2026.01 | 220.4 | 4.5 | 9 | |
| DPVision=ResNet-18×2, Action=UNet, Params=281.19M, Size=~4.4GB, Steps=16 (DDIM)2026.01 | 385.3 | 2.6 | 5 | |
| DPVision=ResNet-18×2, Action=UNet, Params=281.19M, Size=~4.4GB, Steps=100 (DDPM)2026.01 | 2,007.5 | 0.5 | 1 |