Reproduce “Look Before You Act: Bridging Active Perception and Vision-Language-Action Models via Observation-Aligned Data Adaptation” 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

Jiawei Geng, Zhenyang Liu, Yuzheng Wu, Yongchong Gu, Yanwei Fu

ActiveAlign adapts a vision-language-action policy to manipulate after a wrist-camera viewpoint change. It generates training examples by matching active-view endpoints to demonstration states, aligning them with feedback control, and retaining manipulation continuations, without changing the policy architecture. On perturbed-start LIBERO Spatial tests, it reaches 78.2% success versus 54.6% for a matched-state control; physical tests cover three apple tasks and tested occlusions.

Summary by alphaXiv.

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