Reproduce “Red-Teaming of Vision-Language-Action Models via Quality Diversity Environment Generation” 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
Shihan Zhao, Henry Chen, Ya-Chuan Hsu, Minjune Hwang, Siddharth Srikanth, Freddie Liang, Akanksha Saran, Aaquib Tabrez and 2 more
Q-DEG searches simulator scenes that expose failures in vision-language-action models while keeping instructions fixed, using diversity in pre-action policy activations to guide the search. It fine-tunes models on successful trajectories from those scenes. On held-out Q-DEG environments, π0.5 success rises from 38.03% to 81.81%; this result is from the 10 simulated LIBERO-Spatial pick-and-place tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.