Reproduce “Learning Loco-Manipulation From SMPC Demonstrations With Sparse Offline-to-Online RL” 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

Martin Schuck, Maks Sorokin, Simone Manni, Duy Ta, Angela P. Schoellig, Marco Hutter, Simon Le Cleac'H, Jan Brüdigam

The authors use sample-based model predictive control (SMPC) demonstrations to train sparse-reward loco-manipulation policies. In simulation, SMPC supplies successful transitions to an off-policy learner, while a frozen low-level controller adjusts the legs for balance. Across five simulated tasks, expert-data runs reach near-perfect success, while runs without demonstrations fail to learn. Spot box pushing succeeds in 9 of 10 real-world trials.

Summary by alphaXiv.

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