Atlas Steward - Meta

Inspiration

We believe intelligence should serve human flourishing.

Some of the most important decisions people make involve complex systems: water, infrastructure, ecosystems, resources, and communities. Yet these systems are usually represented through flat maps, dashboards, spreadsheets, and reports.

We asked a simple question:

What if you could hold a complex problem in your hands?

Atlas Steward turns a physical table into a living mixed-reality world where people can see relationships, explore interventions, understand trade-offs, and make decisions alongside an AI steward.

Our first experience focuses on a real-world challenge: a neighborhood facing recurring flood risk.

What it does

Atlas Steward is a hands-first spatial decision-making experience for Meta Quest.

Users place a miniature living environment onto their real table through passthrough. They can investigate terrain, roads, drainage, water, buildings, and ecological features using natural hand gestures.

Users can:

• Point to areas they want to understand.
• Grab, move, rotate, and inspect parts of the world.
• Place potential interventions directly into the environment.
• Ask questions using voice.
• Explore consequences and resource trade-offs.
• Save decisions and return to the evolving world.

An AI Steward acts as a reasoning partner. It interprets the current world state, explains evidence, identifies uncertainty, compares possibilities, and proposes next actions without taking the final decision away from the human.

The core loop is:

Observe → Understand → Explore → Decide → Remember.

How we built it

Atlas Steward was designed as a new mixed-reality experience using Unity and Meta XR technologies.

The experience combines hand tracking, passthrough, spatial anchoring, voice interaction, a structured world model, and an agentic intelligence layer.

Instead of putting a conventional chatbot inside VR, we made the AI part of the environment. Its reasoning can change what the user sees, highlight relationships, answer contextual questions, and respond to interventions.

We built the first challenge as a focused, controlled simulation so the experience remains reliable during live judging while demonstrating a scalable architecture for future real-world datasets and scenarios.

Challenges we ran into

Our biggest challenge was resisting the temptation to build too much.

Atlas Sanctum is a much larger vision, but the competition rewards a complete experience that can be understood and enjoyed quickly. We therefore had to reduce a complex decision-intelligence platform to one compelling spatial interaction loop.

We also had to design interactions that remain intuitive without controllers, work within a small seated space, and make AI reasoning feel useful rather than gimmicky.

Finally, we had to balance an open-ended AI experience with deterministic simulation behavior so users can experiment freely without the demo becoming unpredictable.

Accomplishments that we're proud of

We created an experience in which the physical table becomes the interface.

Instead of navigating menus, users manipulate a living system with their hands. Instead of receiving a generic AI answer, they can see the consequences of decisions spatially.

We are especially proud of the central interaction philosophy:

Don't put reality into a dashboard. Put the decision inside the world.

We also designed the architecture so the same interaction model can eventually support water resilience, ecological restoration, agriculture, infrastructure, emergency planning, and other complex systems.

What we learned

We learned that great spatial computing is less about adding more information and more about changing how information is understood.

A flat map can tell you where something is.

A spatial world lets you reason about relationships.

We also learned that AI becomes more compelling when it is grounded in context, uncertainty, and action rather than simply generating text.

Most importantly, we learned that the best role for an AI steward is not to replace human judgment, but to expand the field of view available to the person making the decision.

What's next for Atlas Steward

Atlas Steward is the first interface for a larger Atlas Sanctum vision: an intelligence layer between reality and action.

Next, we want to connect the experience to real geospatial and environmental data, introduce additional worlds and scenarios, improve persistent spatial memory, and allow teams to collaborate around shared decision spaces.

Long term, Atlas Steward could help people and institutions understand complex systems, evaluate interventions, coordinate action, and preserve the memory of what was learned.

Our vision is simple: give people a larger field of view, then let them choose what to build with it.

Built With

• agentic-ai
• ai
• computer-vision
• digital-twins
• environmental-intelligence
• geospatial-intelligence
• hand-tracking
• human-ai-interaction
• meta-quest
• meta-xr-sdk
• mixed-reality
• multimodal-ai
• passthrough
• simulation
• spatial-computing
• spatial-ui
• unity
• voice-interaction
• xr