"Select-to-Act: Hierarchical Reinforcement Learning via Adaptive Language Guidance" proposes a hierarchical policy to explicitly select which language instruction guides the next states before acting under it, rather than aggregating all instructions into one mixed signal. #RL