Industrial Defect Detection & AMR Warehouse System
+25% factory throughput · Production CV inspection + full AMR warehouse automation across 14 months at Vstech
Jul 2024 – Sep 2025
+25%
Throughput Increase
<3 min
Delivery Time
Problem
A manufacturer needed two AI systems delivered simultaneously: an automated vision inspector to replace inconsistent manual defect checking on production lines, and a fully autonomous mobile robot (AMR) for warehouse pick-and-deliver — both shipping to production on a fixed 14-month timeline with no margin for failure.
Approach
Engineered defect-detection CV models (PyTorch/YOLO) on real production-line imagery — replacing subjective manual inspection with consistent, automated quality control across multiple product categories.
Accelerated deployment to production by exporting trained models to ONNX and serving via ONNX Runtime in Docker/Flask microservices — enabling single-command containerized releases and cutting inference overhead vs PyTorch baseline.
Architected a complete AMR warehouse automation system end-to-end: integrated Cartographer SLAM, ROS2/Nav2 localization, and a REST task API to deliver autonomous pick-and-place across the full warehouse floor.
Boosted factory throughput by 25% and cut pick-to-delivery cycle time to under 3 minutes by solving the task-scheduling bottleneck — owning every layer from model training to edge deployment.
Results
Eliminated manual re-inspection bottleneck on production lines — improving quality consistency and freeing operator capacity for higher-value tasks.
Shipped containerized model releases with zero-downtime deployments via Docker + ONNX Runtime microservices.
Delivered a fully autonomous AMR pick-and-deliver workflow covering the entire warehouse floor, running reliably in production.
Achieved +25% throughput and <3-min cycle time — validated on live production data at Vstech over a 14-month engagement.
Architecture Overview
Training (PyTorch/YOLO)
ONNX export
ONNX Runtime
Flask REST API
Docker container
AMR controller (ROS2/Nav2)
Localization
Task scheduler
Warehouse management
How to Run
Build the inference Docker image: docker build -t defect-inspector .Run the Flask service: docker run -p 5000:5000 defect-inspectorPOST /predict with a base64-encoded image to get defect bounding boxes and class scores.For AMR integration, install ROS2 Humble + Nav2 on the robot controller.Launch AMR stack: ros2 launch amr_nav2 bringup.launch.pyTask scheduler runs as a standalone REST service: python task_scheduler.py