Firmware release notes¶
2026.09.10-10hz10ms — bench candidate¶
Download and install this candidate for the horizontal v3 STM32H563RGT6 carrier with eight v2 VL53L8CX sensor boards.
| Setting | Value |
|---|---|
| Sensor grid | 8×8 per sensor; 512 zones total |
| Sensor rate | 10 Hz per sensor |
| Sub-integration | 10 ms, four per 8×8 frame |
| Ranging mode | Autonomous |
| Output | AVOID=cp: MAVLink OBSTACLE_DISTANCE at 10 Hz |
| Zone orientation | 180°, matching the fitted v2 sub-boards |
| Host UART | 921600 baud, 8N1 |
Measurement handling¶
Weak, inconsistent, and no-target readings now leave their direction unknown. A successful sensor read, an unmasked field of view, or a missing echo no longer declares clearance to 4 m. Masked returns, excluded floor/ceiling returns, out-of-range readings, and gaps between measured angular directions also remain unknown.
Accepted status 5 or 9 measurements provide obstacle distances. Status 9 can represent a merged target; it is accepted as an obstacle, not as proof of clearance. Target-count output is enabled so a zone without a target is marked as no data rather than trusting a leftover distance/status.
Close obstacles below 50 cm are retained. The advertised minimum is 2 cm; a positive reading below that is reported at the minimum rather than discarded. Use deliberate zone masks for enclosure and airframe returns.
The avoidance path keeps each sensor's latest frame between polls and expires it 200 ms after its successful polling pass, including when acquisition stalls. New invalid readings replace old valid readings immediately. These host timestamps do not compensate for exposure time, transport delay, or vehicle motion. The raw browser protocol retains its original per-poll fresh sensor bitmap.
Unknown directions can restrict movement
PX4 Collision Prevention may restrict movement into open sky or poorly
measured directions. Do not enable CP_GO_NO_DATA merely to hide missing
measurements. This release does not claim continuous calibrated coverage
between individual zone rays or guaranteed 4 m visibility outdoors.
Power and timing¶
Each complete 8×8 frame uses four sub-integrations. The candidate has a 100 ms frame interval and 40 ms of exposure per frame: 40% calculated exposure duty. This is not total board power.
| Profile | Frame interval | Exposure per frame | Exposure duty |
|---|---|---|---|
| Earlier 15 Hz / 5 ms experiment | 67 ms | 20 ms | 30% |
| This candidate: 10 Hz / 10 ms | 100 ms | 40 ms | 40% |
| Earlier 15 Hz / 10 ms profile | 67 ms | 40 ms | 60% |
Longer exposure can improve weak-target detection, with less frequent updates than 15 Hz / 5 ms. It does not automatically correct calibration errors. Compare input watts, missed detections, and measurement age with the same scene, supply, and peripherals. See Power.
Installation and saved settings¶
Startup and sensor recovery use shared configuration/readback checks for mode, rate, resolution, and integration. The packaged installer checks those values on all eight sensors after programming. Saved zone masks keep their existing layout and orientation, and the installer checks their bytes before/after.
Validation status¶
Passed: ARM compilation/link; obstacle-status, close-range, mask, angular-gap, freshness/expiry, and timer-wrap regressions; existing protocol and orientation tests; 15 installer tests; package-integrity checks.
Pending: an end-to-end hardware installation and all-eight-sensor readback of this exact release. The installation attempt stopped during identification, before backup or flash, because the ST-Link was absent from USB. Earlier 5 ms flashes do not validate this candidate. Windows/Linux hardware installation, power, range, optical calibration, and recovery fault injection remain unverified.
The ZIP includes VALIDATION.md, offline installation instructions, the release
manifest, checksums, and third-party notices. It is a bench candidate, not a
flight-validation result.
References: ST integration timing and target-status interpretation, MAVLink distance and unknown values.