- role
- timeline
- stack
turn iphone lidar scans into geometry an app can actually trust. raw depth is noisy, alignment drifts, and a wrong surface is worse than no surface.
- building spatial capture pipelines on iphone lidar
- aligning depth and mesh data into usable surfaces
- gating output on capture confidence, not hope
- keeping processing on device where it matters
- optimizing geometry under risk, not just fit
capture you can build on
the work is spatial capture and geometry systems on iphone for physical-world applications. the phone scans, and what comes out has to be good enough to drive real decisions downstream.
that sets the bar differently from a demo. every stage of the pipeline has to answer the same question: is this scan actually good enough to use.
evidence first, output second
raw lidar depth is noisy. meshes drift. coverage has holes. so the system treats confidence as a first-class output: depth and mesh alignment get checked, coverage gets measured, and weak captures get flagged instead of quietly passing through.
the rule is simple. nothing downstream should have to guess whether the geometry is trustworthy. the capture either earns its confidence or it says so.
a scan without a confidence story is just a pretty point cloud.
geometry that respects uncertainty
once a surface is captured, the interesting problem is what to compute on it. fitting lines and shapes through noisy real-world data means reasoning about risk, not just best fit: prefer the robust answer over the flashy one, and keep the heavy lifting on device so the loop stays fast and private.
details stay intentionally generic here. the engineering is the point: reliable spatial systems on commodity phones.
- iphone lidar
- arkit · depth
- confidence-gated
- on device
- risk-aware