The Simulator

Walk the block by feel.

Watch a scripted walk, or steer it yourself. Then turn the street off and go by the buzz alone. Eleven scenarios, from a morning sidewalk to a beach run. Three sensors that each miss something the others catch. Every timing here is modeled from the engineering record or is the simulator's own working value, until the bench measures it.

A scripted walk. Zero effort.

IDLE

Street

0.0 / 30 m

Press Play to watch the block by feel.

PACE1.4 m/s

LASER grid, forward

4x4, 60 Hz, 2.85 m

RADAR

8 m, ahead, 1 track

SONAR

2 beams, 66 ms

World to skin

MODELED

about 70ms

AHEAD, 4x4 at 60 Hz, about 10 cm at 1.4 m/s

sensor frame 25
readout 5
fusion 5
drive and rise 20
skin 15

Ground mode about 120 ms. The controller's own share is up to 10 ms ahead and 16 ms for the ground. Modeled from the parts' datasheets until the bench measures it.

Active hits

1 detections

Radar, Person passing, 7.3 m

Waveform

last 2.0 s

Keys, left right or a d steer, space play, r restart, b by feel, m sound, s scenario, Esc picture back

This is a picture of the idea. The first generation is in use since 2022; the second is in build.
World to skin is about 70 ms for a hazard ahead and about 120 ms for the ground, modeled from datasheets until the bench measures it.
Bright sun shortens the laser grid's range. The bench sets the number. Sonar covers large surfaces in that gap; radar covers only what moves.
In this simulator, sonar misses soft targets beyond about 1 m: a simulator assumption to be measured. The record says only that sonar is weak on small or sharply angled targets. The laser grid sees a coat; radar sees the person inside if they move.
The simulator does not model the radar returns the wearer's own walking creates, nor the IMU milestone that trusts the radar less when the wearer stops. The record has both; the bench measures them.
The ground grid here sees a drop off out to 2.6 m, the sonar cone is drawn at 48 degrees, and the sonar's two beam cycle is 66 ms: simulator design values, to be measured.
The walking pace of 1.4 m/s and the running paces are the simulator's working values; the record fixes only about 3 m/s for running and measures walking speed in the tests.
The record has one head channel at the collar strip. The simulator's left, center, and right collar split is its own working choice.
Nobody has proven the stairs coding on real stairs yet. The bench decides.
Street crossings are not solved by any device. The mobility aid and your hearing come first.
Pods on the bands hold the motors, each with a soft face against the skin; there are no adhesives and no pockets.

Why this architecture was chosen, from the archived simulator spec: Katzschmann and colleagues (IEEE TNSRE 2018) put a time-of-flight belt and a haptic strap on 12 blind users for 162 trials, with zero collisions in blindfolded trials. That is another team's result with their own belt, not a result for this kit; this kit's own results come at the milestones. The timings in this simulator are working values until the bench measures them.

This is the kit's language on a screen. The kit itself is in development. On the real kit, about 70 ms pass from the world to your skin for a hazard ahead and about 120 ms for the ground, modeled from the parts' datasheets until the bench measures it. Distance sensors carry the warning, with no camera and no AI in the safety loop. The three laws hold everywhere: where it buzzes is where it would hit you, how strong is how close, and when it moves you follow it.