Add baro temperature to sensor voter observation #9411
Conversation
When determining the confidence of a barometer sensor, we should consider the temperature as well, alongside the pressure. Low-noise baros can show the same pressure reading for a second or two when not moving and in an indoor location.
|
Yes this makes sense, and it's what we've done with the differential pressure sensors. The other thing you could try is inserting the pressure as pascals. Are these repeated samples byte for byte identical floats or is the difference simply less than the hard coded epsilon in the validator? |
|
What board and barometric sensor are you using? |
|
This is good to merge, but I'd still consider the pascal conversion in another PR. |
|
I was not around in the last days, sorry for the late response and thanks for your reply.
Good point, I just checked that. The 24-bit raw measurement is not as stable as the pressure, meaning it would not trigger a "stale" warning. However, the pressure (milibars) that is sent via
It's an mpc2520. I suppose this is the official website: http://www.maiersensor.com/products/air-sensors/mpc2520.html. I heard from my colleagues, that the driver was in master at one point, or at least there was a PR. But apparently we were the only ones using the sensor.
Thanks. Not sure if I understand correctly: Would you then revert this change at a later point? |
For our baro sensor (no driver in px4 master), I keep getting STALE errors every few seconds since a recent merge of master.
The errors appear while the vehicle is steady on a table indoors. Apparently the pressure reading from the baro can stay at a specific value for roughly a second under these conditions. As soon as
> 100(source in ecl) samples have the same value, the error above is triggered and the sensor flagged as stale.A recent PR (#8466) modified the way baro measurements are published. Instead of the derived altitude, the message now only contains the pressure, which makes sense. Before that change however, the altitude was compared in the sensor voter and not the pressure. The altitude computation might have introduced some numerical magic that caused the measurement to be a bit more noisy - numerically speaking. 😕
My idea is to also add the baro temperature to the voter. But I am not sure if that's smart. What if the baro reading actually does become stale, but the temperature reading does not? I don't have enough experience to know what failure scenario the stale check is trying to detect.
@dagar any input? :)
Altitude example output:
Pressure example output of same sensor, different px4 version: