Reproduce “LASER: Learning a Latent Action Space for Efficient 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
Arthur Allshire, Roberto Martín-Martín, Charles Lin, Shawn Manuel, Silvio Savarese, Animesh Garg
LASER learns a task-relevant action representation from robot experience and lets policies explore through actions suited to related tasks. It encodes and decodes robot commands, preserving actions and predicting state changes. In simulated Panda wiping, offline LASER reaches reward 300 in 600,000 policy-training steps, versus 1.4 million in the original action space; Door transfer is less optimal.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.