Goal-conditioned Robot Manipulation on Libero (test)
45Success Rate: Put Red Mug LeftGCP-P
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GCP-Psolver=DDPM, denoising steps=1002026.03 | 45 | 40 | 55 | 30 | 70 | 80 | 65 | 15 | 50 | 10.8335 | 1 | |
| DDIM10solver=DDIM, denoising steps=102026.03 | 40 | 25 | 50 | 40 | 45 | 85 | 50 | 25 | 45 | 1.002 | 10.8119 | |
| GCP*source=original paper2026.03 | 38.4 | 40.8 | 51.2 | 38.4 | 70.4 | 79.2 | 72.8 | 25.6 | 52.1 | — | — | |
| GCP-Isolver=DDIM, denoising steps=1002026.03 | 30 | 40 | 40 | 40 | 50 | 70 | 70 | 15 | 44.375 | 10.8089 | 1.002 | |
| DPMsolver=DPM-Solver, denoising steps=102026.03 | 30 | 40 | 45 | 35 | 50 | 80 | 55 | 5 | 42.5 | 1.004 | 10.7903 | |
| GCP + DTSinference framework=DTS, step size n1=10, step size n2=2, chunk length=62026.03 | 30 | 40 | 70 | 25 | 65 | 90 | 65 | 15 | 50 | 6.7578 | 1.6031 |