ResearchBenchmarksContinual Reinforcement Learning on HumanoidBench (First Five Tasks Sequence Final Checkpoint)Continual Reinforcement Learning on HumanoidBench (First Five Tasks Sequence Final Checkpoint)Follow25Average Success RateSPHERE2.128.061419.94May 6, 2026Evaluation resultsMethodMethodLinksAverage Success RateSPHERE2026.0525Top-K MoE2026.053