Reproduce “Plan, Watch, Recover: A Benchmark and Architectures for Proactive Procedural Assistance” 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
Kaustav Kundu, Ritvik Shrivastava, Maxim Arap, Nanshu Wang, Xianhui Zhu, Quintin Fettes, Gautam Tiwari, Parth Suresh and 9 more
The authors introduce PWR, a procedural assistant that tracks a visual-cue-linked plan and revises it after interventions to support recovery from skipped, reordered, or substituted steps. They evaluate it on Pro²Bench across procedures. Against the same trained models without plan context, PWR improves three aggregate metrics; zero-shot models reach at most 26% of reference interventions before an error. These are frozen evaluations, not live user trials.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.