Collecting real-world data at scale creates a privacy problem that better 3D reconstruction alone cannot solve.
A street capture might include faces, license plates, or details that were never meant to become part of an AI dataset. Those details can be irrelevant to the spatial model but still sensitive to the people being recorded.
That's why the privacy layer matters in
@vangrid_io's approach. Its capture workflow includes masking faces and license plates, helping make crowdsourced spatial data more practical to use.
But automated masking isn't the whole answer. Buyers still need useful geometry, and contributors need clear rules about where filming is permitted and how sensitive footage is handled.
I think privacy protection should be treated as part of data quality, not a feature added after collection.
What should matter more when scaling Physical AI datasets: stricter privacy safeguards, or preserving as much visual detail as possible for reconstruction?