Robot Manipulation Success Rate on LIBERO-10 (Tasks 2, 3, 5, 8, 9 Test Set)
134Success CountWeak First All-Mix New Norm
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Weak First All-Mix New NormStep=6000, Roll-outs=50 per task2026.06 | 134 | 53.6 | |
| Strong First All-Mix New NormStep=9000, Roll-outs=50 per task2026.06 | 123 | 49.2 | |
| Strong All All-Mix New NormStep=5000, Roll-outs=50 per task2026.06 | 121 | 48.4 | |
| Strong First Libero 10 MixStep=7500, Roll-outs=50 per task2026.06 | 118 | 47.2 | |
| Strong First Libero10 Collected MixStep=1000, Roll-outs=50 per task2026.06 | 107 | 42.8 | |
| Weak First All-Mix Old NormStep=500, Roll-outs=50 per task2026.06 | 98 | 39.2 | |
| Weak First Baseline All-Mix Old NormStep=1000, Roll-outs=50 per task2026.06 | 86 | 34.4 | |
| Strong First Only Low LRStep=100, Roll-outs=50 per task2026.06 | 86 | 34.4 | |
| Strong First Baseline All-Mix New NormStep=3000, Roll-outs=50 per task2026.06 | 83 | 33.2 | |
| Original baselineStep=0, Roll-outs=50 per task2026.06 | 80 | 32 | |
| Fisher 1e12 Low LRStep=9999, Roll-outs=50 per task2026.06 | 80 | 32 | |
| Weak First Only Old NormStep=500, Roll-outs=50 per task2026.06 | 10 | 4 | |
| Strong First Only Old NormStep=1000, Roll-outs=50 per task2026.06 | 1 | 0.4 | |
| Strong All Only Old NormStep=1000, Roll-outs=50 per task2026.06 | 0 | 0 |