Spatial Web: Multi-Agent Information Fusion

A Spatial Web architecture for unified observation, feature, and decision fusion across distributed intelligent agents — grounded in IEEE 2874-2025 and built for Earth observation systems that must reason, coordinate, and establish trust across organizational boundaries.

📜 IEEE 2874-2025 Spatial Web 🧠 HSML · HSTP · UDG 🔐 DIDs & Verifiable Credentials
George PercivallSpatial Web Foundation · GeoRoundtable
Stephane FellahSpatial Web Foundation · Geo(k)noesis

Information fusion is central to modern Earth observation. Integrating heterogeneous data sources into actionable knowledge is what turns sensing into decisions. Observation (sensor) fusion and feature fusion are well established within the geospatial and remote sensing communities. Decision fusion — particularly in distributed, autonomous, multi-agent environments — remains insufficiently addressed.

Existing fusion frameworks are largely pipeline-centric and centralized. That design limits their ability to support interoperable reasoning, coordination, adaptation, and trust across heterogeneous agents, platforms, and organizational boundaries. The gap is not in the sensors. It is in the framework that lets independently-owned agents share what they observe, what they conclude, and why they should be believed.

This work proposes a Spatial Web–based architecture that unifies observation, feature, and decision fusion. The approach is grounded in the IEEE Spatial Web Protocol, Architecture, and Governance Standard (IEEE 2874-2025), developed by the Spatial Web Foundation, which defines a semantic, spatial, and governance framework for interoperable interaction among network-connected entities — sensors, IoT devices, autonomous platforms, digital twins, software agents, and humans. It extends established fusion concepts from IEEE GRSS and the Open Geospatial Consortium by embedding them in an agent-centric, semantically grounded, trust-enabled spatial model.

The gap
Decision fusion in multi-agent EO environments is insufficiently addressed. Pipeline architectures cannot support interoperable reasoning, coordination, adaptation, and trust across heterogeneous agents, platforms, and organizational boundaries.
The shift
From centralized data pipelines toward knowledge-driven, agent-centric, trust-aware fusion. Agents reason over shared spatial knowledge, coordinate sensing and analysis strategies, verify the trustworthiness of data and decisions, and collectively generate decisions in dynamic environments.
The mechanism
Three components carry the architecture: HSML for formal semantic representation, HSTP for secure and verifiable agent interaction, and the UDG as the shared knowledge substrate where trusted observations, features, and decisions are published, discovered, and reasoned over with provenance.
Presentation
A version of this presentation including audio track is available as Google Slides: Spatial Web: Multi-Agent Fusion for Earth Observation.
📊 View the presentation

Part 1 · Motivation

The Fusion Challenge

Fusion has a long history in IEEE GRSS and the wider community. Over the years the topic has expanded from data fusion to information fusion — semantically distinct and non-numeric inputs, integrated into actionable knowledge.

Well established

Observation (sensor) fusion

Mature theory and methods, developed within GRSS and the remote sensing community over decades. Multi-sensor, multi-temporal, multi-resolution combination of measurements is a solved engineering problem with a deep literature.

GRSS multi-sensor pixel-level
Maturing

Feature fusion

Benefits from cross-community semantics. Multi-modal and multi-scale features integrate within a common spatial context when they carry standardized semantic representations — the direction set out by Analysis Ready Data principles and FAIR-AI standardization.

ARD FAIR-AI semantics
The gap

Decision fusion

Accelerating rapidly with AI agents — and the least supported by existing frameworks. Bringing sensing, observation, and features into a decision framework that can support or autonomously take action requires agents that can reason, coordinate, adapt, and establish trust across boundaries no pipeline was designed to cross.

AI agents coordination trust
Why now
Decision fusion is accelerating because AI makes possible what was not possible before — reasoning, coordination, and adaptation at machine speed. What is missing is the framework that lets agents coordinate. That framework is what the Spatial Web supplies, and it is what decision fusion for Earth observation now needs.

Part 2 · Architecture

Fusion Based on Space and Time

IEEE 2874-2025 defines a semantic, spatial, and governance framework for interoperable entities of any kind — organized, at its base, by space and time.

The foundation is an IEEE standard. IEEE 2874-2025, the Spatial Web Protocol, Architecture, and Governance Standard, was developed by the Spatial Web Foundation and approved through the IEEE Computer Society. It is deliberately general about what counts as an entity: sensors, IoT devices, EO and GIS services, digital twins, software agents, and humans all participate on the same terms.

The application of the Spatial Web to fusion extends established fusion concepts rather than replacing them — building on the IEEE GRSS fusion literature and the OGC fusion standards studies. What it adds is an agent-centric model with identity and trust built in, and a semantically grounded spatial model that covers all three fusion types across distributed intelligent agents.

Interoperable Spatial Web entities
Sensors · IoT devices · EO and GIS services · digital twins · software agents · humans — all first-class, all addressable, all governed by the same semantic and trust model.
Extends, not replaces
Schmitt & Zhu (IEEE G&RSM, 2016) and Percivall et al. (OGC 10-184) established the fusion frameworks. The Spatial Web extends them with AI agent-centric structure, identity and trust, and a semantically grounded spatial model.
Side-by-side comparison: traditional pipeline-centric fusion producing isolated decisions, versus Spatial Web agent-centric fusion where intelligent agents share a Universal Domain Graph and produce collective intelligence and adaptive action.

Simplified Spatial Web architecture for information fusion. Left: the traditional pipeline takes heterogeneous sources through observation and feature fusion, centralized processing, and out to isolated decisions. Right: the same heterogeneous sources, but intelligent agents — each carrying reasoning, coordination, and trust capabilities — work over a shared Universal Domain Graph holding observations, features, and decisions. The result is collective intelligence and adaptive action rather than an isolated answer.

DimensionPipeline-centric approachSpatial Web agent-centric approach
TopologyLinear, centralized processing chainDistributed agents over a shared knowledge graph
Fusion productsIsolated decisions at the pipeline terminusObservations, features, and decisions as first-class semantic entities
SemanticsImplicit in code and schemasExplicit and formal in HSML
CoordinationPre-orchestrated by a single operatorNegotiated between agents via activities and contracts
TrustCentralized authority; trust in the operatorDIDs and Verifiable Credentials; policy travels with the data
AdaptationRequires re-engineering the pipelineAgents re-task sensing and analysis in response to shared knowledge
BoundariesStruggles across organizationsDomains carry authority, membership, and norms across organizations

Part 3 · Components

HSML, HSTP, and the UDG

Three components make the architecture work: a formal semantic representation, a protocol for trusted agent interaction, and a shared substrate for collective intelligence.

Semantic core

Hyperspace Modeling Language (HSML)

HSML provides a formal semantic representation for observations, features, activities, agents, and decisions within a shared hyperspace graph. Its conceptual model is defined in IEEE 2874. At the base of the framework sit space, time, and hyperspace. Concepts that persist over time within that framework are domains, which carry identity. Change is enacted by agents performing activities. And governance is contextual rather than top-down: credentials, norms, and contracts define what an agent may do within the domain in which it operates.

The conceptual model has been developed into a standalone modeling language on the W3C stack — RDF, RDF-star, OWL, and SPARQL. Hyperspace is the context for every entity: coordinate spaces where elements are related by paths, and discrete topological cell and graph spaces, with category theory providing the means to move between them. Entities follow a holonic model — each is both a whole in itself and potentially part of a higher domain, and may belong to several domains at once.

Fusion, in HSML, is an activity performed by agents. Observations from remote sensing platforms are modeled as spatially grounded semantic entities, so they can be consistently interpreted and reused across agents. Decision fusion becomes an agent-level process in which goals, hypotheses, actions, and outcomes are first-class semantic objects that can be exchanged, compared, and reconciled.

HSML conceptual stack: space, time and hyperspace as the framework for concepts; domains and identity as concepts that persist over time; agents and activities as concepts that describe change; credentials, norms and contracts as concepts for governance.

The HSML conceptual model. Reading upward: hyperspace frames the concepts, domains persist within it, agents and activities produce change, and credentials, norms, and contracts govern what change is permitted.

Identity and trust
Spatial Web IDs are built on W3C Decentralized Identifiers and Verifiable Credentials, so identity and policy travel with the data. Agents form contracts expressed in HSML, allowing them to negotiate and to determine whether an activity has in fact been achieved.
HSTP conceptual model branching into a messaging binding served by MQTT and a request/response binding served by a Web API.

HSTP bindings. One conceptual model, multiple transport bindings — messaging (MQTT) and request/response (Web API).

Protocol

Hyperspace Transaction Protocol (HSTP)

HSTP is the method for passing HSML: a secure, verifiable protocol supporting semantically validated transactions between agents. It ensures interoperability between diverse AI and digital twin systems, and its operations are based on the actor paradigm.

HSTP incorporates W3C DIDs for identity and provides the basis for the Spatial Web Domain Registration Authority. Multiple bindings are anticipated — messaging and request/response are defined today, and an MCP binding is an open question worth pursuing as agent-tooling protocols converge.

Trust without a central authority
DIDs and Verifiable Credentials let agents, data sources, and individual assertions be cryptographically identified, authenticated, and authorized without reliance on centralized trust authorities — the property that makes cross-organizational fusion tractable.

Knowledge substrate

Universal Domain Graph (UDG)

The UDG is a shared, distributed knowledge substrate for trusted multi-agent fusion — an analogy to the World Wide Web, but of entities rather than documents. Trusted observations, features, and decisions are published, discovered, and reasoned over with provenance and accountability.

🕸️
Knowledge graph

An HSML hypergraph of all entities — observations, features, and decisions — that agents can reason over as shared knowledge rather than as separate data holdings.

🏛️
Distributed registries of domains

Domain membership and SWID registries are where authority is vested. Domains define norms and rules, issue credentials, and establish the trustworthiness of data and decisions.

🧭
Hyperspace features

Independently trained world models converge on common features — structurally and semantically isomorphic across models. The AI analog of geo-objects, needing Hyperspace Reference Systems to relate them.

🤝
Social network of agents

Agents coordinate through activities and contracts, jointly setting sensing and analysis strategies. Collective intelligence is the emergent property of that coordination.

A distributed network graph with dense clusters connected by sparse long links, illustrating the social network of agents in the Universal Domain Graph.

The UDG is heterogeneous by design. Dense clusters within domains, sparse links across them — a social network of agents rather than a uniform database.

Independently trained world models converge on common features — isomorphic not only in structure but in meaning. This is what makes World-to-Vec engineerable, and it is what lets agents fuse decisions rather than merely exchange data.

Hyperspace features · Anthropic J-space, Neuronpedia · after Goodchild's geo-objects

World2Vec Hyperspace Reference Systems Collective intelligence

Part 4 · Applications

New Capabilities & Application Domains

What the architecture enables for Earth observation, and where it applies.

New capabilities for Earth observation

01

Scalable collective intelligence

Across heterogeneous platforms and organizations, without a central integrator.

02

Adaptive closed-loop sensing

Agents coordinate sensing and analysis strategies in response to what the shared graph already knows.

03

Interoperable decision-making

Decisions cross agent boundaries as semantic objects that can be compared and reconciled.

04

Verifiable trust

Cryptographic identity and credentials for agents, data sources, and individual assertions.

Application domains

🛰️
Coordinated satellite and airborne observation

Tasking, tipping, and cueing across independently operated constellations and platforms.

🚨
Disaster response

Near real-time integration of observations, features, and decisions, where GeoAI advances already benefit from agentic coordination.

🌡️
Climate and environmental monitoring

Long-horizon, multi-organization observation programs with shared provenance.

🏗️
Smart infrastructure management

Autonomous agents integrating sensing and decisions across owners and jurisdictions.

Conclusion

A Foundation for Next-Generation Earth Observation

🧩
Unified fusion framework

Observation, feature, and decision fusion unified within a common semantic, spatial, and trust-enabled framework — shifting from pipeline-centric architectures to knowledge-driven, agent-centric models.

📜
Grounded in open standards

IEEE 2874-2025 Spatial Web, IEEE GRSS fusion frameworks, OGC standards, W3C DIDs and Verifiable Credentials, and Analysis Ready Data principles for multi-modal feature integration.

🌐
Interoperable and autonomous

Systems spanning heterogeneous agents, platforms, and organizations, with autonomous agents capable of collective and trustworthy reasoning in complex spatial environments with real-time adaptation.

The architectural shift

Heterogeneous sources → Semantic entities (HSML) → Trusted exchange (HSTP) → Shared graph (UDG) → Collective decisions

Key References

Source Material

Standards, fusion literature, and trust specifications underpinning the architecture.

Related Work

Explore Further

GeoRoundtable work on Spatial Web standards, hyperspace features, and agent architecture.

📊 Presentation slides (with audio) 🧭 Spatial Web Features: World-to-Vec Realized 🛰️ Spatial Web Baseline 2025 📜 IEEE 2874-2025 Standard Approval 🧩 UDG Physical Design Synthesis 🌐 Concepts of Space from AI 🔗 Spatial Web Foundation

Build fusion that crosses organizational boundaries

GeoRoundtable brings together expertise in geospatial standards, agentic AI, information fusion, and systems engineering. We help organizations move from centralized pipelines to trusted, agent-centric architectures.

Get in touch