Robotic Manipulation on LIBERO Object
99.4Success RateForgeVLA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ForgeVLAE=100, P=1, # Params (M)=3882.724, # Trainable Params (M)=128.1012026.05 | 99.4 | 100 | |
| Centralized# Params (M)=3882.72, # Trainable Params (M)=128.102026.05 | 98.8 | 100 | |
| CentralizedE=100, P=1, # Params (M)=3882.724, # Trainable Params (M)=128.1012026.05 | 98.8 | 100 | |
| CentralizedE=10, P=10, # Params (M)=3882.724, # Trainable Params (M)=128.1012026.05 | 98.8 | 100 | |
| ForgeVLA# Params (M)=3882.72, # Trainable Params (M)=128.102026.05 | 98.6 | 100 | |
| FedAvgE=100, P=1, # Params (M)=3882.724, # Trainable Params (M)=128.1012026.05 | 98.2 | 100 | |
| FedAvg# Params (M)=3882.72, # Trainable Params (M)=128.102026.05 | 97.6 | 100 | |
| MolmoAct22026.07 | 97 | — | |
| π0.52026.07 | 96 | — | |
| ForgeVLAE=10, P=10, # Params (M)=3882.724, # Trainable Params (M)=128.1012026.05 | 95.6 | 100 | |
| GaP2026.07 | 95 | — | |
| A2A-ExplicitVisual conditioning=A2A-GroundingModel mask rendered as an RGB highlight, Action head=Diffusion Policy (DP) [10]2026.06 | 94.4 | — | |
| FedAvgE=10, P=10, # Params (M)=3882.724, # Trainable Params (M)=128.1012026.05 | 94 | 100 | |
| DP-RGBVisual conditioning=raw RGB observations, Action head=Diffusion Policy (DP) [10]2026.06 | 87.8 | — | |
| π0.5 w/GaPNavigation=GaP, Policy=π0.52026.07 | 85 | — | |
| WAM-RLDescription=Ours2026.06 | 82 | — | |
| πRLDescription=actor-only reinforcement learning2026.06 | 78 | — | |
| A2A-ImplicitVisual conditioning=intermediate features from A2A-GroundingModel, Action head=Diffusion Policy (DP) [10]2026.06 | 75.8 | — | |
| UAD-DPVisual conditioning=fine-tuned UAD [51] affordance heatmap, Action head=Diffusion Policy (DP) [10]2026.06 | 74.4 | — | |
| MolmoAct2 w/GaPNavigation=GaP, Policy=MolmoAct22026.07 | 70 | — | |
| BaseDescription=pretrained WA model without RL2026.06 | 68 | — | |
| TipTop2026.07 | 22 | — | |
| FedVLA* + CLIP# Params (M)=519.82, # Trainable Params (M)=92.902026.05 | 18.2 | 30 | |
| FedVLA*# Params (M)=79.23, # Trainable Params (M)=79.232026.05 | 2.2 | 10 |