Reproduce “Skill Machines: Temporal Logic Skill Composition in 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

Geraud Nangue Tasse, Devon Jarvis, Steven James, Benjamin Rosman

The authors present skill machines (SMs) that compose reusable goal-reaching skills with logical constraints and a finite-state task machine to execute temporal tasks. Constraint tracking lets a skill pursue a goal while avoiding forbidden features. In Office Gridworld, ordinary Q-learning fails the hardest task, while SMs and guided Q-learning learn hard long-horizon tasks, subject to pretrained-skill costs and reachability assumptions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.