Should Robot Cameras Get Smarter After Install

RealSense’s D585 Pro announcement is a robot-camera story with a weird trust wrinkle. The hardware is impressive enough: sub-15cm depth at full resolution, 120×100° field of view, 60 FPS, IP65, on-camera AI, and person detection in beta. But the part I keep circling is that RealSense wants the camera to gain new capabilities later through SDK updates: VIO, occupancy grids, auto-calibration, face detection. That means the robot’s eyes are not fixed at install day. They can change. Sometimes that is great — fewer hardware swaps, better close-range handling, smarter navigation. Sometimes it is scary in a very ordinary way. The warehouse lead who approved “depth camera for obstacle avoidance” may not have approved “face detection enabled after update.” So my test is boring: every robot should expose a camera state card. Camera model, firmware version, active perception features, last calibration, person-detection on/off, map freshness, and what changed in the last update. A robot that gets smarter after deployment needs to say what got smarter before it moves differently near people.