4. Operational Scenarios: From SLAM to AI World Models

SpatialDDS is designed to be practical and flexible across real-world deployments. The following scenarios illustrate how the Core, Discovery, Anchors, and Extension profiles can be combined in different ways to support robotics, AR/XR, smart city, IoT, and AI-driven applications. Each scenario lists the profiles involved and the key DDS topics flowing in and out, showing how the schema maps onto actual use cases. Optional profiles such as Neural and Agent are marked clearly, allowing implementers to see future directions without requiring them in the baseline.

Core SLAM/SfM Scenarios

These scenarios cover the foundational use cases for spatial mapping and localization. They show how devices and services exchange features, images, pose graphs, and geometry tiles to support SLAM and structure-from-motion pipelines, either on-device, at the edge, or in multi-agent systems.

  1. On-device Visual(-Inertial) SLAM
    A single device runs its own SLAM, fusing camera and IMU, publishing nodes/edges.
  2. Profiles: Core (Pose Graph, VIO)
  3. Topics In: raw IMU/camera

  4. Topics Out: pg.node, pg.edge, geo.tf

  5. Device → Edge Distributed SLAM
    A mobile device streams features/images to an edge server for map building.

  6. Profiles: Core, SLAM Frontend
  7. Topics In: feat.keyframe, BlobChunk (images)

  8. Topics Out: pg.node, pg.edge, geom.tile.*

  9. Multi-Agent SLAM with Global Alignment
    Multiple devices contribute to a shared map, aligning through anchors.

  10. Profiles: Core, Anchors
  11. Topics In: pg.node/edge from peers

  12. Topics Out: pg.edge (loop closures), geo.tf (frame alignment)

  13. Offline SfM / Batch Reconstruction
    A service reconstructs geometry from stored images/features.

  14. Profiles: Core, SLAM Frontend
  15. Topics In: BlobChunk (image sets), feat.keyframe
  16. Topics Out: geom.tile.meta/patch/blob

Service Scenarios

These scenarios describe how SpatialDDS supports services that go beyond local SLAM, such as Visual Positioning Services (VPS), cooperative relocalization, map delivery, and anchor registries. Discovery messages and manifests play a key role here, allowing clients to find and interact with services dynamically.

  1. Relocalization / Place Recognition
    A service matches incoming features against a prior map for relocalization.
  2. Profiles: Core, SLAM Frontend
  3. Topics In: feat.keyframe

  4. Topics Out: geo.fix, pg.nodegeo

  5. VPS — Features-only Query
    Client sends features; service returns a pose or node with geo anchor.

  6. Profiles: Core, SLAM Frontend
  7. Topics In: feat.keyframe

  8. Topics Out: geo.fix or pg.nodegeo

  9. VPS — Image-only Query
    Client sends an image; service extracts features and returns pose.

  10. Profiles: Core
  11. Topics In: BlobChunk (role=“image/jpeg”)

  12. Topics Out: geo.fix or pg.nodegeo

  13. Cooperative VPS / Crowd Relocalization
    Devices share queries and matches to improve coverage.

  14. Profiles: Core, Discovery
  15. Topics In: feat.keyframe, geo.fix

  16. Topics Out: shared pg.edge or consensus geo.fix

  17. Mapping Service Consumption (Discovery)
    Clients discover and fetch map tiles for their area of interest.

  18. Profiles: Core, Discovery
  19. Topics In: disco.service, geom.tile.meta/patch/blob

  20. Topics Out: local cache of geometry

  21. Anchor Registry Subscription (Discovery)
    Clients subscribe to a registry of persistent anchors.

  22. Profiles: Core, Anchors, Discovery
  23. Topics In: anchors.set, anchors.delta
  24. Topics Out: geo.tf (local → anchor/world alignment)

Consumer Scenarios

These scenarios focus on AR clients and applications that consume maps, anchors, and semantics. They show how SpatialDDS delivers persistent content alignment, semantic overlays, and shared localization for end-user experiences.

  1. AR Client Map Consumption
    An AR headset consumes geometry and anchors to render content.
  2. Profiles: Core, Anchors
  3. Topics In: geom.tile.*, geo.anchor, geo.tf

  4. Topics Out: none

  5. Semantics-Assisted Mapping
    A client enriches tiles with object detections for smarter AR overlays.

  6. Profiles: Core, Semantics
  7. Topics In: geom.tile.blob

  8. Topics Out: semantics.det.3d.set

  9. AR Client with VPS + Anchor Registry
    A client uses VPS fixes plus anchors for persistent localization.

  10. Profiles: Core, Anchors, Discovery
  11. Topics In: feat.keyframe or image blobs, anchors.set
  12. Topics Out: geo.tf

Lifecycle / Recovery Scenario

This scenario illustrates how a device or client can quickly catch up with the current state of the world after joining late or recovering from a failure. By fetching cached tiles and anchors, clients can synchronize efficiently without disrupting live streams.

  1. Catch-Up & Recovery (Reality Feed Style)
    A late joiner fetches cached tiles/anchors to sync quickly.
  2. Profiles: Core, Anchors
  3. Topics In: geom.tile.meta/patch/blob, anchors.set
  4. Topics Out: resumed pg.node/edge

AI & World-Model Extensions

These scenarios extend SpatialDDS beyond SLAM and AR into the realm of AI agents, neural maps, and digital twins. They demonstrate how AI perception services, planning agents, and predictive twin backends can plug into the same bus, consuming and enriching the shared world model. Neural and Agent profiles are optional extensions, and scenarios that use them are marked accordingly.

  1. VLM/Detector as a Perception Service
    An AI model consumes images and publishes 2D/3D detections.
  2. Profiles: Core, Semantics (+ SLAM Frontend if features in)
  3. Topics In: geom.tile.blob, optionally feat.keyframe

  4. Topics Out: semantics.det.2d.set, semantics.det.3d.set

  5. Captioning / Visual QA Agent
    A vision-language model provides captions/labels tied to anchors or tiles.

  6. Profiles: Core, Semantics
  7. Topics In: geom.tile.blob, geo.anchor

  8. Topics Out: semantics.det.2d.set (with captions), agent.answer (optional)

  9. Neural Map — Remote View Synthesis (optional Neural extension)
    Thin clients request rendered views from a neural map service.

  10. Profiles: Core (+ Neural if adopted)
  11. Topics In: neural.view.req

  12. Topics Out: neural.view.resp with images

  13. Neural Map — Asset Streaming (optional Neural extension)
    Neural assets (e.g., Gaussian splats) are streamed as blobs for local rendering.

  14. Profiles: Core (+ Neural if adopted)
  15. Topics In: geom.tile.meta/patch (encoding=“nerf”/“gaussians”)

  16. Topics Out: none

  17. Digital Twin Ingest (Realtime → Twin Backend)
    A digital twin backend ingests SpatialDDS streams for persistent modeling.

  18. Profiles: Core, Semantics, Anchors
  19. Topics In: pg.node/edge, geom.tile.*, geo.anchor, semantics.det.3d.set

  20. Topics Out: none

  21. Digital Twin → SpatialDDS (Predictive Overlays)
    A twin service publishes predictions or overlays back to clients.

  22. Profiles: Core, Semantics
  23. Topics In: none (internal twin logic)

  24. Topics Out: semantics.det.3d.set, geom.tile.*

  25. Route/Task Planning Agent (optional Agent extension)
    An AI agent consumes world state and publishes goals or routes.

  26. Profiles: Core (+ Agent if adopted)
  27. Topics In: pg.node, geo.tf, semantics.det.3d.set, geo.anchor

  28. Topics Out: task.route, task.goal, or agent.task/status

  29. Human-in-the-Loop Labeling & Training Data Capture
    Detections are corrected by humans and fed back for model improvement.

  30. Profiles: Core, Semantics
  31. Topics In: geom.tile.blob, semantics.det.* (proposals)
  32. Topics Out: semantics.det.* (corrected), data.capture.meta

Taken together, these scenarios show how SpatialDDS functions as a real-time bus for spatial world models. From raw sensing and SLAM pipelines to AR content, digital twins, and AI-driven perception and planning, the protocol provides a common substrate that lets diverse systems interoperate without heavy gateways or custom formats. This positions SpatialDDS as a practical foundation for AI world models that are grounded in the physical world.