Exploiting Local Dynamics Regularity for Reusable Skills in Offline Hierarchical RL

In three linesCARL, an offline hierarchical reinforcement learning algorithm, exploits local dynamics regularity to discover reusable skills. By aligning global contexts with required action sequences, the method improves performance on OGBench when integrated with HIQL.

Summary generated by Claude — human-verified