Computer vision · Human–AI interaction
I’m Xia, a researcher working as a software engineer at Google. I build AI that connects an understanding of the physical world with the people who live and act in it.
Now at Google / Software Engineer Computer Vision & Spatial AI
Build spatial representations that connect what AI perceives with what people know.
Explore FlyMeThroughA conceptual view of my research
01 / Selected research
The same room can offer different possibilities—and present different barriers—to different people. My research asks how AI can understand those differences and help us act on them.
SPATIAL REASONING
Can an AI understand where you can go?
A benchmark for capability-conditioned indoor navigation: evaluating whether vision-language models reason about routes in relation to an agent’s physical capabilities.
Benchmarking Vision Language Models on Capability-conditioned Indoor Navigation
PERSONALIZED ACCESSIBILITY Watch film
Accessibility starts with the person.
Personalized scans of built environments, connecting visual observations to an individual’s accessibility needs.
Personalized Accessibility Scans of Built Environments
HUMAN–AI COLLABORATION Watch film
A map is more than its geometry.
Human–AI collaborative indoor mapping with commodity drones, bringing human knowledge into reconstructed 3D spaces.
Human-AI Collaborative 3D Indoor Mapping with Commodity Drones
SITUATIONAL AWARENESS Watch film
Make barriers visible. Make spaces better.
An augmented reality tool for scanning rooms and identifying potential accessibility and safety issues in the built environment.
Room Accessibility and Safety Scanning in Augmented Reality
SPATIAL AUTHORING Watch film
Turn visual understanding into creative control.
A design workflow that combines depth estimation, semantic understanding, and geometry extraction to author 2.5D designs.
Authoring 2.5D Designs via Depth Estimation, Semantic Understanding, and Geometry Extraction
CONTEXT-AWARE INTERACTION Watch film
What should an augmented world sound like?
Context-aware sound generation that connects virtual interactions with the real-world environments in which they take place.
Context-aware Sound Generation in Augmented Reality
02 / Current work
How can AI understand people and spaces together, within real-world resource constraints?
At Google, I explore how computer vision algorithms and agents can work across on-device and cloud systems. The goal is a layered understanding of environments, the people within them, and how their actions relate to spatial context.
I’m interested in turning that understanding into timely, context-sensitive assistance—including surfacing potential safety risks in response to people’s needs, while working within limited computational resources.
03 / A continuing thread
Xia Su Researcher & engineer
My path began in architecture. The question that stayed with me: how can spaces work better for the people who use them?
At the University of Washington’s Makeability Lab, advised by Jon E. Froehlich, my doctoral research brought that question into human–AI interaction, computer vision, and accessibility.
I think of this as human–AI symbiotic spatial perception: combining AI’s ability to perceive and reason with people’s knowledge, capabilities, and intentions. Today, that perspective guides my work on spatial intelligence at Google.
View my full CV PDF04 / Let’s connect
I welcome conversations with researchers and teams working on spatial AI, human–AI interaction, accessibility, and intelligent systems in the physical world.
xiasu@cs.washington.edu