I joined Esri in 2011 to address facilities and indoor mapping challenges, which led to the creation of ArcGIS Indoors. Since late 2021, I have been leading Esri’s Innovation Lab, a team that operates between product development and delivery—beyond traditional roadmaps and backlogs. This position enables us to examine complex customer workflows that remain unresolved in ArcGIS and to investigate transformative technologies to bridge those gaps.
One of our most persistent challenges has been helping customers quickly build adaptable systems of record that can serve as the foundation for a digital twin. The difficulty of creating reliable representations of the built environment on campus, in buildings, and indoors largely arises from traditional reliance on AEC project documentation. So, what’s the problem?
– As-built conditions often don’t match original plans.
– BIM models, while powerful, are often too complex for day-to-day operational use.
– CAD files are often poorly structured, while PDFs are unstructured, making integration into GIS difficult.
A Paradigm Shift: Reality Capture + GeoAI
Our research has revealed a paradigm shift in how detailed and accurate “building information” can be generated through a combination of Reality Capture (RC) and GeoAI, enabling these data to provide customers with digital twin capabilities readily. In some respects, this feels like a “back-to-the-future” moment.
Early in my career, in the Army and while managing the US Coast Guard’s base mapping program, we relied on satellite and aerial data to capture reality snapshots. We then used a combination of manual digitization and early machine learning to extract planimetric features—roads, fence lines, and buildings. The accuracy was only as good as our sensors and ground control, but the fundamental principle was clear: the best source of truth is the real world.
Fast forward to today, and we’re witnessing a similar trend. Except now, instead of satellite imagery, we’re utilizing consumer-grade reality capture devices like GoPro 360s, iPhones with LiDAR, and commercial drones.
The Breakthrough: GeoAI-Driven Indoor Mapping
Imagine sending a barely trained person to walk around a campus with a GoPro 360. Using GeoAI, that simple walk-through can generate high-fidelity 3D models of buildings and interiors. But it doesn’t stop there—GeoAI can then automatically extract, classify, and structure the features inside those models into a spatially aware 3D GIS repository.
The floorplans extracted from these AI-driven models are often more accurate than CAD or BIM files.
We validated this capability in a recent research project. A customer had previously captured LIDAR and panoramic imagery across millions of square feet of facilities to create BIM models. Later, when they required a comprehensive, spatially enabled security asset database, they encountered a lengthy manual effort to document every camera, sensor, and access control point.
Instead, we used GeoAI to analyze the existing data—extracting features from LIDAR depth and applying AI-based depth estimation to identify and classify security assets. The result? What would have taken years was automated in a fraction of the time.
Prediction and Pattern Recognition with GeoAI
Two of the most powerful capabilities of GeoAI are predictive spatial analysis and pattern recognition—both critical to automating geospatial workflows.
In our Gaussian splat pipeline, for example, we leverage radiance fields to predict and reconstruct missing frames, which allows us to fill in the gaps in captured data. This ensures smooth, accurate digital twins without requiring exhaustive, 100% coverage scanning. The same principles apply at larger scales—whether predicting infrastructure changes over time or reconstructing incomplete building models with AI-driven inference.
GeoAI is also fundamentally changing how we detect and extract key features. Rather than blindly searching an entire dataset, we use models like CLIP to narrow the universe of images to search, prioritizing the most relevant data before extraction begins. Whether identifying specific building assets or classifying security infrastructure, this targeted approach reduces noise and increases accuracy, making feature extraction scalable and precise.
The New Value Proposition: GIS as the Source of Truth
All of this raises a fundamental question:
Why not use GIS to update and create CAD & BIM rather than vice versa?
We’re now in a world where it’s easier than ever to:
– Generate high-accuracy, photo-realistic 3D models of your buildings and interiors using Gaussian splatting and advanced GeoAI techniques.
– Extract the spatial features you need automatically using GeoAI-driven recognition and depth estimation.
– Turn those insights into a dynamic, operational GIS, serving as a system of record and a digital twin.
GeoAI saves time—it automates what used to take weeks or months. It enables living systems of record to continuously evolve with the real world rather than remain static representations.
The Future of Facility Management: AI-Powered Digital Twins
This convergence of AI, GIS, and Reality Capture fundamentally changes how organizations interact with their built environment. Instead of relying on outdated documentation, we can create spatially intelligent, continuously updated digital twins that reflect reality.
The implications are enormous:
> Speed and Scale: Tasks that once required manual effort can now be automated at scale, unlocking new levels of efficiency.
> Enhanced Decision-Making: AI-powered insights enable faster, more informed decisions based on real-world conditions.
> Collaboration and Cloud Integration: GeoAI workflows support cloud-based collaboration and seamlessly integrate with industry-standard tools like NVIDIA Omniverse and other 3D ecosystems.
The Big Shift: From Documentation to Intelligence
A GeoAI-enabled GIS allows our users to leap beyond the foundational capabilities of a system of record, allowing them to create digital twins of their facilities and business operations. The future isn’t just about visualizing facilities and business operations—it’s about leveraging the digital twin to understand and interact with them in real time to drive better outcomes in the real world.
The question isn’t whether AI will transform geospatial workflows. It already has.
The real question is: How will you leverage GeoAI and ArcGIS to redefine what’s possible?
I’m presenting in the Revolutionizing Construction Visualizations and Analyses with GeoAI, Radiance Fields, and Gaussian Splatting [S71236] session at NVIDIA GTC, pop in and say hi!
