Robotic Manipulation on PushT (test)
89Success RatePRISM-MPPI (ours)
Evaluation results
| Method | Links | |
|---|---|---|
| PRISM-MPPI (ours)Learned-prior params=1.0M, Sample budget (K)=1282026.06 | 89 | |
| PRISM-MPPI (ours)Learned-prior params=1.0M, Sample budget (K)=642026.06 | 86 | |
| PRISM-MPPI (ours)Learned-prior params=1.0M, Sample budget (K)=322026.06 | 82 | |
| Vanilla MPPI (LeWM)Learned-prior params=0, Sample budget (K)=642026.06 | 61 | |
| Vanilla MPPI (LeWM)Learned-prior params=0, Sample budget (K)=322026.06 | 59 | |
| Vanilla MPPI (LeWM)Learned-prior params=0, Sample budget (K)=1282026.06 | 57 | |
| Diffusion PolicyLearned-prior params=19.3M, Sample budget (K)=K-independent2026.06 | 41 | |
| BC-only (head µ alone)Learned-prior params=1.0M, Sample budget (K)=K-independent2026.06 | 31 | |
| PRISM-MPPI (DINO-WM-style)Learned-prior params=1.0M, Sample budget (K)=1282026.06 | 15 | |
| PRISM-MPPI (DINO-WM-style)Learned-prior params=1.0M, Sample budget (K)=322026.06 | 13 | |
| PRISM-MPPI (DINO-WM-style)Learned-prior params=1.0M, Sample budget (K)=642026.06 | 10 | |
| Vanilla MPPI (DINO-WM-style)Learned-prior params=0, Sample budget (K)=1282026.06 | 5 | |
| Vanilla MPPI (DINO-WM-style)Learned-prior params=0, Sample budget (K)=322026.06 | 4 | |
| Vanilla MPPI (DINO-WM-style)Learned-prior params=0, Sample budget (K)=642026.06 | 3 |