Bidirectional DShot telemetry probabilistically fails to start on some AM32 ESCs

[mbuzdalov]

Bug Report

With current master, ESC entries don't start coming in at arming in one of my copters, which prevents further spooling up via TKOFF_RPM_MIN. This happens with certain noticeable probability (ranging from 0.25 to 0.75), and apparently does not go away with repeated arming attempts, so power-cycling is needed.

This is bisected to commit fab4b5e with a change relevant to DShot ESCs.

Bug report

Issue details

Please describe the problem

Version

Reproducible starting with commit fab4b5e.
Does not happen with the previous commit b6b4b26.

Potentially relevant configuration features: fast rates are enabled, SERVO_DSHOT_ESC set to 3 "BLHeli32/AM32/Kiss+EDT". Parameters are attached to the issue.

Platform
[ ] All
[ ] AntennaTracker
[x] Copter
[ ] Plane
[ ] Rover
[ ] Submarine

Airframe type
Quad (seemingly irrelevant).

Hardware type

Flight controller: FlywooF745 version 3.
ESC firmware: AM32 2.20.

Logs

Logs are rather irrelevant. ESC entries simply don't start coming.

Parameter dump is attached below.
2026-03-02-0retune-again.txt

[LupusTheCanine]

I can confirm that this happens on heli too, thankfully all motors are affected simultaneously.

[andyp1per]

We should probably back out @rishabsingh3003 's change for now

[rmackay9]

It must be this 2 second delay or something in the call to "read_radio" that is required for some reason to initialise DShot. RC input affecting output is very strange though.

In terms of timing, we have perhaps 2 days before we branch for 4.7. I'm aiming for this Friday...

[mbuzdalov]

It is interesting that @LupusTheCanine reports the issue on a Heli. The commit I have bisected to apparently saves up just a single parameter read in the case of Heli. (Unless I misunderstand something, on Heli this routine never performs this radio wait, so this may probably not be responsible for the issue.)

For any other issues, I can (at the level of my knowledge of Ardupilot, of course) only imagine issues with:

• rpmdata.last_update_us from the AP_ESC_Telem subsystem having a special value of 0;
• various possible wraparound issues involving this value (e.g. I find it sketchy how rpmdata.update_rate_hz updates and consequently how get_rpm works);
• and maybe the fact that telemdata.last_update_ms is also often interpreted as 0 meaning no data, but it is not set accordingly (e.g. telemdata.last_update_ms = AP_HAL::millis();).

These issues above might be transient in the normal use, but probably not during startup, where something may become zero in too delicate moments (probably quite an issue with the last point), and which may possibly lock the system up in an unusable state. The fact that we arrive to this code earlier in the startup with fab4b5e may increase the probability of nasty things happening. The current state feels a bit too complicated for the present me to exclude the latter with confidence.

UPD: the above is apparently unrelated to this issue, as additional investigations by inserting some logging in AP_ESC_Telem::update() before any filtering happens shows that no telemetry is coming in from the backend (as in: all last_update_* are always zeros).

[rishabsingh3003]

I will try to reproduce this locally on a BLHeli32 ESC because I don't have AM32s.

[LupusTheCanine]

@mbuzdalov It might have been a fluke or another issue, now I am unable to reproduce now. One thing I am certain about is that we don't exit passthrough mode the way AM32 expects as I have to power cycle ESCs after programming.

[mbuzdalov]

It is very reproducible on my side, although, as I mentioned, this happens with a probability. Sometimes I have to power-cycle four to five times before it agrees to fly, sometimes it requires to power-cycle three times for the telemetry to fail. Quite a signature of timing issues, to my taste.

Another thing, which occasionally happens subject to the problem manifesting is the ESC initialization delay, is that some of the ESCs beep much more times than usual. This happens much less frequently than the main issue (no ESC telemetry).

I recall that many months ago I occasionally had missing ESC telemetry as well, which was the reason to enable TKOFF_RPM_MIN, as the lack of ESC telemetry means very bad flight on this machine.

Maybe I'm just the only unlucky guy.

[LupusTheCanine]

False alarm on heli.