Reproduce “Direction-Conditioned Policies for Online Goal-Conditioned Reinforcement Learning” 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
S K Swaminathan, Damiya Gondha, Theyanesh Eswaramoorthy Rajahkrishnan, Aritra Hazra
DCP gives a goal-reaching policy a learned direction and distance to its target rather than raw goal coordinates. It selects a visited waypoint during training, then uses the final goal directly at deployment, without waypoint selection or planning. Across nine tasks, DCP exceeds CRL in final success and time near the goal on seven each, but loses on both measures in AntSoccer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.