Reproduce “PLAS: Latent Action Space for Offline 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
Wenxuan Zhou, Sujay Bajracharya, David Held
PLAS learns offline reinforcement-learning policies by training a deterministic policy to choose latent actions, which a pretrained conditional variational autoencoder decoder converts into environment actions. Bounding latent outputs is intended to favor dataset-like actions, without guaranteeing support. On D4RL HalfCheetah medium-expert, unperturbed PLAS reports average reward 11,716.9 versus BCQ’s 7,750.8; this comparison does not establish an advantage on every dataset.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.