How DePIN Powers Location for Physical AI

Learn how DePIN networks combine RTK, visual positioning, and spatial computing to give robots the real-world location awareness needed to operate across farms, cities, and indoor environments.

Robots are no longer confined to factories and warehouses. They are making their way into farms, construction sites, city streets, and even our homes. To navigate and function safely in these ever-changing environments, they require precise awareness of their location (positioning) and a clear understanding of their surroundings (mapping). While GPS offers a solid foundation, it falls short in delivering the accuracy and reliability needed for such diverse, dynamic settings

This challenge is being met with new solutions built on Decentralized Physical Infrastructure Networks (DePIN). By combining community-driven networks, vision-based positioning, and collaborative spatial computing, DePIN is providing a new foundation for robotic navigation.

The Importance of Location Services for Physical AI

Artificial intelligence is extending beyond the digital realm and into the physical world, integrating with the physical world by blending robotics, spatial awareness, and an understanding of physics into what’s now known as Physical AI. 

This emerging field enables systems to interact with their environments through physical interfaces—whether robots, sensors, or distributed networks—rather than remaining confined to virtual computations. Currently, Physical AI thrives in controlled settings, such as factories or labs, where variables can be tightly regulated and outcomes reliably predicted.

To effectively navigate outdoor environments, Physical AI will require robust, real-world data drawn from dynamic, unpredictable situations—far what simulations or controlled datasets can provide.

Humans rely on vision and an intuitive understanding of their surroundings to move through the world. In recent years, we’ve augmented this capability with smartphones equipped with GPS and mapping services to enhance navigation and access information outside our immediate perception. This adoption of digital technology marked the first step in augmenting reality with data not readily visible to the naked eye. Augmented Reality (AR) builds on this by overlaying digital information onto our physical field of view. With the advent of smart glasses, AR will soon enrich our understanding of where we are and what we see in real time. 

physical ai

Yet, while humans will benefit from this technology, robots are likely to emerge as the primary users of location-based metadata, blending visual data with additional contextual layers—such as spatial coordinates, environmental conditions, and object recognition. For a robot, there is fundamentally no distinction between the sensory data it captures from its immediate environment and the locational data it receives; both are simply streams of information to be processed and acted upon.

Three Pillars of Location-Aware Robotics

To enable Physical AI to thrive outdoors, we’ve identified two critical elements:

  • RTK-based Network: Real-Time Kinematic (RTK) positioning elevates GPS accuracy to the centimeter level, a necessity for robots navigating dynamic landscapes where standard GPS falls short.
  • Visual Positioning System (VPS): By leveraging cameras and computer vision, VPS allows robots to pinpoint their location through visual landmarks, complementing or replacing GPS in obstructed or indoor environments. As part of this system, augmented reality (AR) metadata provides robots with enriched contextual information, enhancing decision-making and adaptability.

vertical farm physical ai location aware robotics

RTK-Based Networks

For Physical AI to function effectively outdoors, centimeter-level precision is critical. Standard GPS, which has an accuracy of a few meters, is insufficient for use cases like precision agriculture, autonomous construction, or robotic delivery services.

GEODNET is building the world’s largest decentralized Real-Time Kinematic (RTK) network to improve location accuracy. By early 2025, it will have more than 13,500 ground-based reference stations across 142 countries, providing centimeter-level precision. These stations use triple-frequency, multi-constellation antennas to correct satellite signal errors caused by atmospheric interference, orbit shifts, and signal reflections. The adjustments are then sent in real time via protocols like NTRIP to devices such as drones and autonomous robots.

Unlike traditional RTK systems, which depend on expensive, centralized infrastructure, GEODNET takes a decentralized approach. It encourages individuals and organizations to set up GNSS “satellite miners” through blockchain-based incentives. This method allows for a faster and more widespread network. Supporting GPS, GLONASS, Galileo, and BeiDou, with reference stations typically within 20-40 kilometers of users, GEODNET provides reliable real-time corrections. It is also developing hybrid PPP-RTK methods and monitoring space weather to refine atmospheric models for even greater accuracy.

For Physical AI, this level of precision is key. It allows drones to land exactly where they need to and robotic planters to sow seeds in straight, even rows. With open access, compatibility with most RTK receivers, and a transparent station map, GEODNET is making high-accuracy positioning available to more users in real-world environments.

“GEODNET is delivering centimeter-level accuracy to thousands of automated drone flights each day through customers like Rock Robotic, DroneDeploy, and Propeller Aero. We’re also providing RTK services to mobile ground robots, including self-driving vehicles, robotic lawnmowers, and robotic dogs—machines performing repetitive, monotonous tasks traditionally handled by humans.” - Mike Horton, Project Creator of GEODNET."

- Mike Horton, Project Creator of GEODNET

Visual Positioning Systems (VPS)

Understanding a robot’s location is critical for applications like package delivery drones or food service robots navigating busy sidewalks, where precise positioning is necessary. A Visual Positioning System (VPS) uses cameras, computer vision, and machine learning to analyze surroundings—landmarks, objects, or pre-mapped features—determining a device’s exact location and orientation. Like human visual navigation, VPS excels where GNSS falls short: indoors, underground, or in dense urban areas.

delivery drone physical ai

Companies like Google (via ARCore) and Niantic have pioneered this technology, with Niantic aiming to construct a machine-readable, 3D map of the world. Collecting the vast, detailed data required for VPS is challenging without compelling incentives or widespread participation, challenges Niantic has cleverly overcome thanks to Pokémon Go. Through the popular game, Niantic taps into player activity, particularly optional AR scans of PokéStops and other landmarks, to crowdsource 3D spatial data. This gamified exchange, offering entertainment for user-contributed scans, has fueled Niantic’s VPS, enabling precise augmented reality experiences and laying the groundwork for advanced spatial intelligence applications.

Crowdsourcing high-value data from engaged contributors is a key feature of Decentralized Physical Infrastructure Network (DePIN) projects. Hivemapper, for example, has successfully built an alternative to Google Street View, mapping roads with impressive accuracy. This provides a strong model for VPS innovation. However, while services like Hivemapper focus on vehicle-based mapping, pedestrian pathways and indoor spaces remain largely uncharted. 

In the Web3 ecosystem, projects like Auki Labs, Meshmap, Over the Reality AI, and Frobobots AI are tackling these gaps with distinct approaches, such as augmented reality (AR)-driven spatial mapping and AI-powered indoor navigation. These initiatives offer promising avenues for expanding VPS capabilities beyond outdoor roads. XMAQUINA should investigate these projects to assess their potential for integration, collaboration, or inspiration in building a more holistic decentralized mapping infrastructure.

Indoor Positioning and Collaborative Spatial Computing 

For indoor applications, technologies such as Ultra-Wideband (UWB), Bluetooth Low Energy (BLE), and magnetic positioning are more precise than satellite-based systems. These solutions, commonly known as Indoor Positioning Systems (IPS), provide accurate location tracking in environments where GPS signals are unreliable or simply unavailable, such as inside buildings but they are very hard to deploy and not scalable. Visual Positioning Systems can play a role also for indoor positioning. 

The Auki network introduced a new approach with its "posemesh"—a decentralized machine perception network and collaborative spatial computing protocol. Designed to enable digital devices to share spatial data and computational resources, Auki facilitates the creation of a collective, real-time understanding of the physical world among connected systems, enhancing their ability to navigate and interact indoors.

Auki also enables the persistence of digital information anchored to physical locations, allowing devices to tag, retrieve, and interact with location-specific data seamlessly. This capability supports applications like augmented reality overlays, robotic coordination, or smart infrastructure management, where precise, persistent spatial awareness is essential.

Nils Pihl, founder and CEO of Auki Labs, highlights the indispensable role of spatial computing in advancing Physical AI:

“We know that 70% of the world economy is still tied to physical labor and physical space, so teaching AI to understand our world is absolutely crucial. Unlocking interoperable, permissionless collaborative spatial reasoning is the holy grail of physical AI.”


He further elaborates on the importance of interoperability and open standards in this transformation:

“Robots perceive the physical world in an entirely digital way through its cameras and other sensors. The challenge is translating the physical world into digital information - but the opportunity is that once this is done well, robots and other devices can collaboratively reason about the world, becoming interoperable and intercognitive in a way that humans are not.
We recognize that 70% of the world economy is still tied to physical labor and physical space, so teaching AI to understand our world is absolutely crucial. Unlocking interoperable, permissionless collaborative spatial reasoning is the holy grail of physical AI.”

In 2025, GEODNET is partnering with robot manufacturers and developing devices that combine RTK capabilities with vision-based sensors such as LiDAR.

By leveraging GEODNET’s expertise in absolute positioning, these devices can autonomously generate and contribute high-fidelity PoseMesh domains. This enables spatially-aware environments that support less-equipped robots and sensors, advancing collaborative machine perception and reinforcing the infrastructure needed for scalable, decentralized spatial computing.

Location Services as the Backbone of Physical AI

Robots will first move seamlessly between factory floors and open fields, and then later from urban streets to indoor spaces. But this will require massive improvements in location services. 

RTK networks like GEODNET deliver unmatched outdoor precision, VPS empowers vision-based navigation in challenging settings, and collaborative frameworks like Auki’s posemesh unlock indoor potential with a layer of critical data helping robots perform their tasks effectively.

Together, these technologies—strengthened by the scalability of DePIN—are shaping a future where autonomous systems can move through the world with confidence and accuracy.

Bullish on robotics? So are we.

XMAQUINA is a decentralized ecosystem giving members direct access to the rise of humanoid robotics and Physical AI—technologies poised to reshape the global economy.

Join thousands of futurists building XMAQUINA DAO and follow us on X for the latest updates.

Owner: