Robotic Manipulation on Place Cube (MA2 problem) 1.0 (real-world)
91.67Success RateTF-DP
Evaluation Results
| Method | Links | |
|---|---|---|
| TF-DPDescription=Full TF-DP incorporating both execution traces and the trace-focused field2026.02 | 91.67 | |
| TF-DP (trace)Description=TF-DP with execution trace only2026.02 | 75 | |
| Gaze Model (Pi-0)Steps=40,0002026.03 | 44 | |
| Base Model (Pi-0)Steps=40,0002026.03 | 32 | |
| DPDescription=Diffusion Policy, Conditioning=None2026.02 | 16.67 | |
| DP-HistActDescription=DP with action history conditioning, History Window=8 past observations2026.02 | 8.33 | |
| Gaze Model (Pi-0)Steps=20,0002026.03 | 6 | |
| Base Model (Pi-0)Steps=20,0002026.03 | 4 |