Continual Reinforcement Learning on HumanoidBench CRL
53Mean ScoreSPHERE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SPHERE2026.05 | 53 | 0.07 | 52 | 0.11 | |
| Top-K MoE2026.05 | 36 | 0.05 | 36 | 0.05 | |
| SPHEREPPO setting=matched Top-K MoE2026.05 | 0.54 | — | — | — | |
| Spectral RegularizationPPO setting=matched Top-K MoE2026.05 | 0.48 | — | — | — | |
| Spectral NormalizationPPO setting=matched Top-K MoE2026.05 | 0.41 | — | — | — | |
| ReDoPPO setting=matched Top-K MoE2026.05 | 0.41 | — | — | — | |
| Top-K MoEPPO setting=matched Top-K MoE2026.05 | 0.36 | — | — | — |