Day–night image pairs as a safety signal
The same street earns very different scores at 2 p.m. and 10 p.m. Why averaging them hides the thing you care about, and how we built the paired-evaluation pipeline.
Notes
Engineering posts, field stories, and essays — written as the work happens. First posts are being drafted now; the topics below are what's coming.
The same street earns very different scores at 2 p.m. and 10 p.m. Why averaging them hides the thing you care about, and how we built the paired-evaluation pipeline.
Street-level imagery alone misses geometry — kerb heights, canopy, overhead wires. How one backbone learns from all three modalities, and where it still fails.
We route on distance because that was the 1990s answer. A founder essay on what an honest navigation system would optimize for instead.
Want these in your inbox when they land? Follow us on LinkedIn ↗ or ask for the email list.