LG AI Research Unveils Autonomous Vision Inspection Automation Framework Using VLM and Agentic AI

LG AI Research has released an autonomous operation framework combining VLM and Agentic AI to resolve Human-in-the-Loop(HITL) bottlenecks in vision inspection systems within smart factories.

Existing systems had structural limitations where engineers had to manually select uncertain data, label it, and retrain the model when Concept Drift occurred. The new framework achieves Zero-Human Intervention from data refinement through retraining, simulation validation, and field deployment by having an LLM Agent control the entire workflow.

The core operating principle consists of four stages. First, boundary cases are automatically extracted via Test-Time Uncertainty Estimation, and the VLM performs Multimodal Auto-Labeling by comparing them with normal reference images. Next, the model is trained to withstand labeling noise by applying a Noise-Robust Learning Policy, and finally, Agentic AI orchestrates the entire pipeline to autonomously deploy the model.