Reproduce “IndusAgent: Agentic Reinforcement Learning for Cross-Category Industrial Anomaly Detection” 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
Rongbin Tan, Fangfang Lin, Jinguo Luo, Min Qiu, Kejin Cui, Mengmeng Wang, Yi Wang, Jiyuan Wang and 4 more
IndusAgent lets an 8B vision-language policy inspect industrial images by requesting crops or visual opinions, then integrating returned observations into a diagnosis. Its accuracy-gated training rewards localization, type, and tool use only when the final diagnosis is correct. Across five benchmarks, the hybrid system reaches 83.5% mean balanced accuracy versus 79.4% for direct Qwen3-VL-Max inference; not a standalone 8B result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.