cd ~/

Persistent Spatial Maps: Remember Where You Were

Designing map persistence systems that let AR content stay in place across sessions - the foundation of spatial computing.

Evyatar Bluzer
2 min read

Every time you put on an AR headset, it wakes up in a new universe. Without persistent maps, the virtual furniture you placed yesterday is gone. Your carefully positioned sticky notes have vanished. AR becomes a toy instead of a tool.

The Persistence Problem

Requirements:

  1. Relocalization: Recognize you're in a previously mapped space
  2. Alignment: Align current tracking to the stored map
  3. Consistency: Virtual content appears in exactly the same physical location
  4. Evolution: Handle environments that change over time

Map Representation

What do we store?

Sparse map: Keyframes with visual features and 3D landmarks.

  • Compact (MBs per room)
  • Fast to load and match
  • Limited to areas with good visual features

Dense map: Full 3D mesh or voxel grid.

  • Complete geometry
  • Large (100s of MBs per room)
  • Better for occlusion and physics

Hybrid: Sparse for localization, dense for rendering/interaction.

  • Best of both worlds
  • Complexity in keeping them synchronized

We're pursuing hybrid - sparse map is the "skeleton" for tracking, dense mesh attached for content interaction.

Relocalization Pipeline

When the headset wakes up:

1. Capture initial frames
2. Extract features
3. Query map database (visual vocabulary / learned descriptors)
4. Candidate map retrieval
5. Feature matching against candidates
6. Geometric verification (PnP + RANSAC)
7. Pose refinement
8. Confidence check → Relocalized!

Target: relocalize in under 2 seconds with >95% success rate in mapped areas.

Map Updates

Environments change:

  • Furniture moves
  • Lighting changes
  • Renovations

Options:

  • Immutable maps: Never update, accept drift
  • Full replacement: Re-map from scratch periodically
  • Incremental update: Merge new observations into existing map

Incremental is ideal but complex. How do you merge conflicting observations? When is a change permanent vs temporary?

We're starting with immutable maps and explicit "remap" user action. Incremental updates are a future optimization.

Storage and Retrieval

At scale, users will have maps of home, office, friends' houses, coffee shops. Potentially hundreds of maps.

Architecture:

  • Local storage for frequently visited places
  • Cloud storage for full history
  • Smart prefetch based on location/calendar

Privacy implications are significant. Maps contain detailed 3D models of private spaces. Encryption, access control, and user consent are paramount.

More on the privacy architecture next month.

Comments