Tag
Proposes GenDa, a unified framework for unsupervised reinforcement learning that addresses non-stationary skill semantics and brittle generalization via skill relabeling and a complementary information bottleneck, significantly improving data efficiency and generalizability.