Reproduce “ReShoot: Generative Visual Domain Randomization of Recorded Robot Demonstrations for Visuomotor Policy 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

Chiyoung Kim, Min Sung Choi, Jinho Ju, Chanhoe Gu, Donghwan Hwang, Wonseok Choi, Woongsun Jeon, Minhyeok Lee

ReShoot augments recorded robot demonstrations by generating videos with changed appearances and matching instructions, while retaining their action labels. An equal-mixture policy combines recorded and generated episodes. It scores 96.5% on standard LIBERO and 85.5% on the sampled LIBERO-Plus evaluation, versus 82.3% for recorded-only data, while synthetic-only training performs worse. The result is limited to tested benchmarks and color shifts.

Summary by alphaXiv.

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