Design of a Smooth Dynamic Digital Twin for Industrial Object Tracking

We discuss the design of a Dynamic Digital Twin (DDT) that provides a photorealistic representation of an industrial facility in real-time. Key challenges include defining an efficient data format for transmitting the state of the facility to the twin, generating an accurate rendering of the detected objects, training a reliable object detection model, and smoothing the rendering so that the representation is not unduly affected by missed detections. We illustrate our DDT in a controlled lab environment with 10 boxes, and report on the smoothness of the representation as boxes are moved around the shelves.