Choosing the sensors (and the hardware around them)
Sensor selection went through four research passes — sunlight alternatives, discreet ultrasonics, an exhaustive sweep, and a final pre-purchase head-to-head — before settling into a staged plan. The guiding rule throughout: no upgrade until a logged test result demands it.
The plan
| Stage | Sensor | Cost | Role |
|---|---|---|---|
| Prototype 1, indoors | Grove VL53L0X (laser ToF) | €8.83 | Desk proof of the haptic interaction |
| Prototype 1, outdoors | DFRobot A02YYUW (waterproof ultrasonic) | €20.90 | All-weather outdoor primary |
| Prototype 1, outdoors | JSN-SR04T V3.0 (waterproof ultrasonic) | €19.51 | A/B baseline and spare |
| Outdoor phase, later | Benewake TF-NOVA (line-beam LiDAR) | ~€55 | Dry-weather thin-branch specialist |
| If size matters | Benewake TFS20-L (1.35 g) | ~€45 | Miniature option |
| Optional experiment | HLK-LD303 24 GHz radar | ~€10 | Sealed-in-case radar channel |
Field data decides the final architecture — one sensor alone, or fusion where an alert requires two sensors to agree (the pattern WeWALK and DJI both use, and the strongest false-alarm defence known).
The head-to-head that picked the outdoor primary
The two waterproof ultrasonic probes look similar on paper. Verification before ordering found they aren’t. The A02YYUW wins on every axis that matters on a cap: a claimed 3 cm blind zone versus the JSN’s independently measured 23–25 cm (uncomfortably close to the 30 cm detection floor), a 60° beam versus 75° (less ground and shoulder clutter), a 30 cm cable versus 2.5 m of captive coil, genuine DFRobot QC versus a clone lottery. Both were bought anyway (€40 total) — the JSN is the A/B baseline, and its years of community documentation are worth having.
The physics behind that blind-zone difference is a nice lesson in itself: single-transducer probes both shout and listen with the same element, which keeps ringing for a millisecond or two after the ping — deaf to close echoes. Two-element sensors listen with a separate, quiet ear.
Why no soldering
Prototype 1’s real question is whether haptic proximity feedback is useful and comfortable — not whether electronics work. So the build is an Arduino UNO R4 Minima with a Grove base shield: every connection clicks in. €35 of core electronics, four boards, nothing to short, no battery handling. Everything learned (firmware logic, vibration patterns, test protocol) carries forward to any later hardware unchanged.
A fancier directional design — multizone sensor, three motors for left/centre/right — was on the table early and got deliberately deferred. It needs soldering and custom circuits, and the single-sensor concept is unproven. It only gets built if field logs show that a single-point warning works but direction is the missing piece.
Rejected, so it doesn’t get re-researched
MaxBotix MB7375 (a tank-level sensor whose 0.5–1.525 m ranging window doesn’t even cover the needed band, ~€120), HC-SR04P (not waterproof), MB1240 (excellent engineering, wrong price and mounting), TF-Luna (superseded by TFS20-L), Broadcom AFBR-S50 (sunlight champion, no Arduino path, $119), the VL53L7/L8CX multizone chips (still sunlight-limited), and TDK’s CH-201 — the MEMS ultrasonic inside the WeWALK — which is buyable as a bare module but has no hobbyist breakout in existence and needs a ~€210+ dev kit. Deferred as an endgame part.
Three bench questions nobody has published answers to, anywhere: ultrasonic echo strength from bare 1–4 cm twigs, line-beam LiDAR on dark branches against full-sun sky, and radar returns from foliage. Prototype 1’s test logs will be original data on the first; the others wait their turn.