Harnessing 3D visuals to create a digital twin of a warehouse for near real-time optimization, operational precision, and resource utilization.
Abstract: The need for precise environmental modeling in warehouses is critical for optimizing freight placement, minimizing downtime, and maximizing resource utilization. This project creates a digital twin of a warehouse to enable near real-time monitoring and validation by comparing Deep Learning approaches, such as Gaussian Splatting and Neural Radiance Fields (NeRFs), with classical techniques like Truncated Signed Distance Functions (TSDF).
A modular pipeline was developed to streamline 3D scene generation, allowing flexible integration of different methods for reconstruction. This modularity supports validation by replacing components and analyzing their impact on scene accuracy and environmental change detection. Deep learning techniques excel in capturing complex geometries, while classical methods provide computational baselines for efficiency.
Reconstructed data, including void detections, is transmitted to a remote server, allowing users to visualize and optimize freight placement. This approach underscores the value of digital twins in achieving operational precision and resource efficiency in dynamic environments.
Advisor: Professor Nitin Sanket
Committee: Professor Loris Fichera, Professor Nitin Sanket, and Mr. Harsh Kakashaniya
Zoom link: https://wpi.zoom.us/j/99178841304