Robotic Manipulation on Real-world Piper-arm tasks (Task Success Rates)
0.9Open Microwave Success RateA2A-Explicit
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| A2A-ExplicitVisual conditioning=A2A-GroundingModel mask rendered as an RGB highlight, Action head=Diffusion Policy (DP) [10], Expert demonstrations=100, Evaluation rollouts=10 per task2026.06 | 0.9 | 0.8 | 0.7 | 0.2 | 0.65 | |
| DP-RGBVisual conditioning=raw RGB observations, Action head=Diffusion Policy (DP) [10], Expert demonstrations=100, Evaluation rollouts=10 per task2026.06 | 0.8 | 0.6 | 0.6 | 0.2 | 0.55 | |
| A2A-ImplicitVisual conditioning=intermediate features from A2A-GroundingModel, Action head=Diffusion Policy (DP) [10], Expert demonstrations=100, Evaluation rollouts=10 per task2026.06 | 0.8 | 0.6 | 0.5 | 0.2 | 0.525 | |
| UAD-DPVisual conditioning=fine-tuned UAD [51] affordance heatmap, Action head=Diffusion Policy (DP) [10], Expert demonstrations=100, Evaluation rollouts=10 per task2026.06 | 0.7 | 0.4 | 0.4 | 0.1 | 0.4 |