Crowdsourced Mapping: Scaling Through Users

Professional mapping doesn't scale. There are millions of places people want AR experiences; we can't map them all ourselves. Users must contribute.

The Vision

Any Quest user can contribute to mapping:

1. Opt in to contribution program
2. Device captures during normal use
3. Captures uploaded (privacy-preserved)
4. Aggregated into maps
5. Maps available for VPS

Everyone benefits from collective mapping.

Privacy Architecture

Users contributing data need strong privacy guarantees:

On-Device Processing

Raw images never leave device:

• Extract features on-device
• Remove identifying information (faces, screens)
• Compress to minimal representation

Differential Privacy

Can't determine if specific user contributed:

• Aggregate contributions before storage
• Add noise to prevent individual identification
• Minimum contribution thresholds

User Control

• Clear opt-in (not default)
• Contribution visibility (see what you've shared)
• Revocation (delete your contributions)
• Geographic limits (don't map near home)

Quality Challenges

User contributions vary wildly:

• Motion blur from casual capture
• Poor lighting conditions
• Obstructed views
• Duplicate coverage

Quality pipeline:

1. Filtering: Reject unusable captures
2. Scoring: Rank captures by quality
3. Selection: Choose best for mapping
4. Validation: Verify accuracy before publishing

90% of contributions are filtered out. The 10% are gold.

Incentive Design

Why would users contribute?

Intrinsic: Help build the AR future Social: Recognition for contributions Utility: Better maps where you care about Gamification: Achievements, progress bars

Avoid:

• Payment (attracts gaming)
• Mandatory contribution (ethics)

Coverage Strategy

Crowdsourced maps complement professional maps:

• Professional: High-value, high-traffic locations
• Crowdsourced: Long-tail, user-specific locations

User contributions fill gaps professionals can't reach economically.

Results So Far

Beta program (10,000 users):

• 1M+ contributions received
• 50K passed quality filter
• 100 new locations mapped (that we couldn't otherwise)
• User satisfaction: 78% would recommend

Scaling to general availability in 2023.

Learnings

1. Privacy concerns are real: Many users opt out due to privacy worry. Must earn trust.

2. Quality bar matters: Low-quality maps hurt more than help.

3. Feedback loops work: Users who see their contributions used contribute more.

4. Long-tail is long: Most contributions are from few users. Engagement is challenging.