Robot Manipulation on LIBERO Spatial (Learning Efficiency Metrics)
85.8Average Success Rate (AVG)Multitask
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MultitaskOracle=true2026.03 | 85.8 | 28.9 | — | — | 51.2 | -0.3 | |
| Sequential Fine-Tuning2026.03 | 81.2 | 24.3 | 0.3 | 3.9 | 57.1 | 5.6 | |
| Expert Replay2026.03 | 80.2 | 23.3 | 0.6 | -2.3 | 49.2 | -2.3 | |
| Dynamic Weight Expansion2026.03 | 79.6 | 22.7 | 0 | 0 | 51.5 | 0 | |
| Dark Experience Replay2026.03 | 73.4 | 16.6 | 4.7 | 0.7 | 55.2 | 3.7 | |
| SLCALR Strategy=Layered LR2026.03 | 69.9 | 13 | -0.6 | 1.5 | 56.1 | 4.6 | |
| Elastic Weight Consolidation2026.03 | 66.1 | 9.3 | 0.7 | 1.5 | 52.6 | 1.1 | |
| RETAINStrategy=Weight Merging2026.03 | 66 | 9.1 | 2.9 | 1.4 | 53.7 | 2.2 |