Reproduce “InteractVLA: Turning VLAs into Interactive Tools for Agentic Robotic Manipulation” 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

Chen Zhao, Zipeng Ye, Shifeng Bao, Yihan Lin, Haitao Shen, Yining Luo, Yuchen Xie, Jing Zhang

InteractVLA lets an agent steer a frozen vision-language-action policy by specifying an end-effector pose, while the policy chooses when to reach it and returns a motion proposal for inspection. Soft-time alignment distributes guidance across possible arrival steps. In online evaluations, success rose from 93.1% to 98.3% on LIBERO-Plus and from 18.0% to 26.0% on RoboDojo, across their reported benchmark conditions.

Summary by alphaXiv.

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