Decentralized Rendering, Physical Retail AI, and the Galbot G1

This week’s update focused on the practical realities of deploying spatial AI in the physical world. From scaling our decentralized reconstruction network to integrating world-class robotics hardware, we are building the perception layer that makes machines useful in retail.

Expanding the Robotics Roster

Building reliable spatial computing requires testing against the best hardware available. Next week, the Galbot G1 arrives at our lab—a highly capable wheeled manipulator from one of the finest robotic engineering teams in China, known for their live retail vending demonstrations.

We are also bringing online a slate of new hardware to ensure the posemesh is robust:

• Unitree R1 and the Unitree Go2 W (wheeled dog).
• A wheeled manipulator from Realman.
• A new, highly affordable robot from a San Francisco-based hardware partner whose core KPI is executing the Auki use case flawlessly.

"Can we do the Auki use case well?" This is the question pragmatic hardware manufacturers are starting to ask as they look to deploy into real business environments.

The Splatter Node: Decentralizing Gaussian Splats

Our reconstruction pipeline takes raw video and spatial data captured in the field and turns it into a unified 3D representation. This week, we launched the Splatter Node, which processes these reconstructions using Gaussian Splat rendering to create a highly detailed, human-readable 3D format.

This is a live, decentralized compute network.

• Consumer GPU Friendly: Splat rendering breaks down into bite-sized chunks (minutes to tens of minutes). A decent gaming rig can process these jobs during idle time.
• Economics: It costs 0.2 credits (20 cents) to turn a scan into a splatter. Reconstructing a supermarket costs around $10.
• Network Growth: We already have over 10 reconstruction nodes and multiple Splatter nodes live, mostly community-hosted.
• Frictionless: The latest update removes the need for static IPs or port forwarding.

"What the reconstruction server does is that it is a bigger brain with more time getting to think about the same data that the robot saw."

State of AI in Retail: From Pilots to Deployment

Fredrik Welin, Head of Customer Success, presented takeaways from NVIDIA’s State of AI in Retail and Consumer Packaged Goods, 2026 Trends. The market has transitioned from the experimental phase to real-world deployment.

A critical gap remains: Retailers have mountains of data (sales, inventory), but lack the spatial context to execute on it.

"Data only creates value when it's connected to action. In the store, location is the central value key."

This is what Cactus solves. It connects retail data to the physical reality of the store—the exact aisle, fixture, and shelf.

• Store Analytics: Made spatial and actionable.
• Staff Optimization: Navigational routing and task efficiency.
• Inventory Management: Pinpointing stockouts to their precise physical location.

As Jensen Huang noted, the next leap is "physical AI"—AI that understands and operates in the real world. For retail, AI must become store-aware. We are already deploying the digital twins that make this possible.

Join the conversation and build the real-world web with us over on Discord.

Watch the whole update on X.

About Auki

Auki is making the physical world accessible to AI by building the real world web: a way for robots and digital devices like smart glasses and phones to browse, navigate, and search physical locations.

70% of the world economy is still tied to physical locations and labor, so making the physical world accessible to AI represents a 3X increase in the TAM of AI in general. Auki's goal is to become the decentralized nervous system of AI in the physical world, providing collaborative spatial reasoning for the next 100bn devices on Earth and beyond.

X | Discord | LinkedIn | YouTube | Whitepaper | auki.com