Reproduce “Privileged Sensing Scaffolds 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
Edward S. Hu, James Springer, Oleh Rybkin, Dinesh Jayaraman
Scaffolder trains robot policies using privileged observations unavailable at deployment, while policies act from target sensors. It adds a privileged world model, exploration policy, and representation-learning objective to improve training alongside a target world model. Across ten simulated tasks, it bridges 79% of the normalized performance gap between target-only and privileged-input DreamerV3, without demonstrating real-robot deployment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.