Quest Pro Launch: VPS in Mixed Reality
Launching VPS as part of Quest Pro's mixed reality system - integration challenges, launch metrics, and lessons.
Quest Pro ships next month with VPS integrated into its mixed reality system. This is our biggest launch yet.
What's Shipping
VPS in Quest Pro means continuous localization for MR experiences, sub-meter outdoor accuracy, integration with Guardian and spatial anchors, and developer APIs for location-aware apps. Coverage at launch is 1,000+ outdoor locations globally, weighted toward urban areas, landmarks, and campuses. Indoor is limited to enterprise partner locations for now.
Integration Journey
Three problems dominated the integration work. VIO integration was hard: merging VPS with the existing tracking stack meant coordinate frame alignment, timing synchronization, and handling the failure modes where the two disagree. Power came next - continuous VPS was too power-hungry as first built, so we added adaptive query frequency, more efficient feature extraction, and smart triggering that skips queries when the system is already confident. And latency: 500ms was not good enough, which drove on-device map caching, predictive prefetching, and an optimized network protocol.
Final Performance
| Metric | Target | Achieved |
|---|---|---|
| Accuracy | under 1m | 0.7m median |
| Latency | under 200ms | 180ms |
| Power | under 100mW | 85mW |
| Success rate | >90% | 88% |
Close on every metric, short on one: success rate came in at 88% against a >90% target. We decided that was good enough to ship.
Against the requirements I wrote in March, the six-month plan held, and latency and power landed where I said they would. Accuracy is the row to read twice: I wrote that MR needs 2cm for believable placement, and the launch target above says under 1m. The 2cm did not stop being true. It stopped being the launch bar.
What could go wrong?
Coverage gaps, first. Many places users will try are unmapped, so the messaging is clear about where VPS works and the system falls back to GPS + VIO where it does not. Accuracy variability, second - some conditions are still challenging, so apps get a confidence indication and anchored content does not display when the pose is uncertain. And privacy: MR raises new questions, which we answer with clear privacy controls and bystander indicator lights.
Developer Readiness
50+ launch partners have VPS-enabled experiences ready: museum tours with contextual information, outdoor AR games with persistent content, campus navigation apps, enterprise training applications. Their feedback shaped the final APIs, which is how it should work.
Post-Launch Plan
The first days are telemetry monitoring, fast-follow bug fixes, and partner support. Over 30 days we expand coverage based on usage data, optimize performance against field metrics, and fold in developer feedback. By 90 days the focus shifts to what comes next: indoor, glasses, general availability for the crowdsourced mapping beta, and Quest 3 preparation.
Personal Reflection
Two years from joining Meta to this launch. In June 2021 I gave consumer-ready quality and core Quest integration twelve months; it took sixteen. VPS went from concept to a shipping product in front of potentially millions of Quest Pro users. The team delivered and the technology works. Whether users care is the one variable we cannot engineer.