Reproduce “Failure Detection for Surgical Robot Imitation Policies via Flow-Matching World Modeling” 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
Zhefeng Huang, Yilin Cai, Ankit Patel, Mohammad Hajiha, Brendan Browne, Yue Chen
FoMo-FD introduces a framework for detecting low-level execution failures in surgical robot imitation policies by modeling nominal visual dynamics using flow matching. The method learns from only successful demonstrations and achieved a 96.6% failure detection rate with a 1.3% false alarm rate on surgical manipulation tasks using a wrist-camera view.
Key insights
- Inverse-transport nonconformity scoring of realized visual transitions provides a more discriminative and reliable signal for window-level consistency than direct prediction error.
- Action-conditioned, window-level dynamics modeling is more effective at discerning low-level execution failures than assessing individual frames for anomalies.
- The wrist-camera view, due to its proximity to the instrument tip, is crucial for detecting fine-grained manipulation failures and significantly outperforms a broader fixed-camera view.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.