Reproduce “Action-Sufficient Goal Representations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jinu Hyeon, Woobin Park, Hongjoon Ahn, Taesup Moon
The authors train an actor-based goal representation for offline hierarchical control, where a planner proposes subgoals and a controller acts on compressed goals. The representation targets action-relevant information rather than only goal value. Actor-based control performs better even with oracle subgoals; on cube-quadruple-play, HIQL-OTA reaches 0.28 success versus 0.01 with its value representation, a result limited to this tested task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.