Consumer tech brand hits 94.7% on VR gesture model

Overview

Spatial VR doesn't ship until the gesture model can tell you what your hands are doing.

A Global Consumer Tech Company preparing a flagship spatial VR headset for global launch needed gesture intelligence that would hold up across lighting conditions, hand-object scenarios, and 20 markets ” with the demographic balance to avoid disparity failures on day one.

Firstsource delivered a labeled gesture corpus with diverse participants from 20 countries ” clearing 94.7% top-1 accuracy and a disparity gap under 1.8% across the launch markets.

This was Intelligence that Operates: gesture data captured for production XR, not for a research prototype.

Challenges

• A spatial headset launches once. The gesture model has to be ready for all 20 markets simultaneously. Sequential market rollouts buy time but cost credibility ” and a model that performs unevenly across demographics is a brand problem, not a metric problem.
• XR gesture recognition isn't standard hand-tracking. Edge-case hand poses, overlapping hands, lighting variability, and real-world hand-object interactions all need representation. Vanilla gesture corpora won't get the model past acceptance testing.
• Demographically balanced training data at this scale doesn't exist off the shelf. Production-grade XR models need data engineered for fairness from the start ” not patched with bias mitigation after deployment.

How We Made It Happen

We built the gesture corpus and the annotation pipeline together so the model cleared launch acceptance and demographic disparity goals in the same program.

• Labeled gesture clips captured for production XR. Diverse hand poses, complex interactions, and real-world conditions ” built for the model the headset would actually ship with.
• Physical AI delivered with demographic balance by design. Participants across 20 markets recruited and captured so the dataset itself carried the diversity the model needed.
• One quality discipline across 20 markets. Corpus and annotation ran under consistent protocols so the model cleared demographic disparity goals in the same program.

Conclusion

Spatial VR ships when the gesture model works for everyone, everywhere, on day one. Firstsource ran the corpus and annotation program to launch readiness ” turning XR gesture intelligence into Intelligence that Operates.