Detecting drop-offs and steps with a single down-facing LiDAR

Cameras are good at naming things and bad at measuring them. For the hazards that injure people — a curb, a staircase, an open manhole, a metro platform edge — Echo 1 uses geometry, not vision: a TF-Luna LiDAR mounted facing down and forward.

The floor is the baseline

On startup, the device auto-calibrates the distance to the floor from the wearer's height and posture. From then on, every reading is compared against that baseline with slow drift tracking, so natural changes in posture don't trigger alarms.

• Reading much farther than baseline → the ground ahead falls away → 'Stop. Drop-off ahead.'
• Reading much closer than baseline → something rises from the ground → 'Caution, step up ahead.'

Killing false positives without missing real hazards

A single anomalous reading is never enough to speak. A confirm-streak filter requires several consecutive out-of-band readings before an alert fires, which suppresses sensor noise, swinging clothing, and momentary tilts. The thresholds are deliberately asymmetric: we treat a missed drop-off as far worse than an unnecessary warning.

Why not ultrasonic sensors?

Ultrasonic rangefinders are cheaper but wide-beamed and slow; they blur a curb edge into a gradual slope. The TF-Luna's narrow optical beam over UART at 115200 baud gives crisp distance transitions — exactly what edge detection needs.