AI is reshaping both how 3D environments get built and how people work with them once they exist. This block opens with a new method for fully automated mobile mapping that removes the need for ground control points, registering aerial and street level imagery to generate millions of control points, then producing seamless Gaussian splats and optimized lidar layers that stream into an enterprise GIS for automatic feature extraction. It closes on the other side of that pipeline, where prompt driven spatial agents let users create splats from image captures and simply ask for encroachment risks, measurements, or extracted objects.
3:30 PM – 4:00 PM
Reality Capture From Air to Ground With Imagery and Lidar – Driving Multi-Scale Digital Twin
Presented by Konrad Wenzel, Esri
Digital Twins perceive reality through sensors. Imagery and lidar acquired from mapping platforms such as crewed aircraft, drones, and ground vehicles enable continuous monitoring of the environment. This includes monitoring assets in cities, facilities, and utility networks, enabling living Digital Twins. At the same time, data capture operations are often limited by manual effort—especially when generating and measuring control for ground-level data. For example, mobile mapping frequently requires ground control points every few hundred meters or lacks the ability to provide reliable accuracy estimates, limiting consistent alignment with other GIS layers in 3D. Accuracy is also critical for multi-temporal analysis, ensuring that objects can be re-identified at the same location over time.
In this presentation, we introduce a new method specifically designed to overcome these limitations and enable fully automated mobile mapping processing without requiring ground control points. Novel AI methods are trained and applied to register aerial and street-level imagery, introducing millions of control points despite significant differences in perspective, appearance, and resolution. A joint bundle adjustment consistently achieves high absolute accuracy while supporting distributed processing of large-scale projects. Lidar point clouds, captured simultaneously by the mobile mapping system, are optimized based on this adjustment.
Gaussian splats are generated from the resulting adjustment using imagery from both domains without visible seams, while leveraging lidar information for highly efficient reconstruction and extraction. During this presentation, the method will be demonstrated along with representative results. Gaussian splats and corresponding 3D layers will be shown integrated into an enterprise GIS platform capable of streaming the data for visualization. Furthermore, automatic feature extraction will be demonstrated, enabling applications such as autonomous asset monitoring.
4:00 PM – 4:30 PM
Talk to Your Gaussian Splat: Prompts for Constructing and Understanding Reality Capture
Presented by Eman Barhoumeh, Ydrive, Inc
Gaussian Splatting has made it easier than ever to create beautiful, highly detailed 3D models of buildings, cities, and infrastructure. However, for most teams, these models still require specialized tools, manual inspection, and considerable effort to extract actionable insights. This session explores an emerging paradigm: automatically transforming image captures into interactive 3D Gaussian Splat environments using simple prompts. Instead of navigating complex software, users can now start asking questions such as: “Create a Gaussian Splat from these images,” “Identify areas where vegetation is encroaching on power lines,” or “Extract the machinery on this factory floor as separate 3D models.” These interactions point to a new layer—often referred to as “Spatial Agents”—where AI systems interpret 3D environments and can help surface insights such as measurements, risks, or areas of interest. Using cross-industry examples, we will explore how prompt-driven interaction could transform 3D models from static visualizations into more conversational systems—and what this could mean for the future of Digital Twins.
