IndusAgent: Reinforcing Open-Vocabulary Industrial Anomaly Detection with Agentic Tools

Researchers introduced IndusAgent, a tool-augmented agentic framework for open-vocabulary industrial anomaly detection.

It combines structured training data, external visual tools, prior retrieval, and gated reinforcement learning to improve classification, localization, and reasoning.

Across benchmarks like MVTec-AD, VisA, MPDD, DTD, and SDD, it aims to reduce hallucinations and strengthen zero-shot generalization.

The approach advances industrial inspection by making multimodal large language models more reliable for anomaly detection and localization.