Near crash: RTL didn't slow down to LAND_SPEED

AC3.2 is fantastic! Very smooth flight, so happy I upgraded from Naza.
I did experience an abnormal behaviour on a TBS Discovery PRO today:

After manually triggering RTL (@line 146200), the quad reached home and began the descent, but strangely “forgot” to maintain LAND_SPEED of 40 cm/s below 10m.

Quickly switched to Stabilize to give throttle just before it would have crashed:

Log seems to show it miscalculated it’s altitude, as both the GPS and baro alt indicate a value 3.5m lower than CTUN Alt - could this be the cause?

I then retried RTL. It climbed and reached RTL_ALT of 20m. After unsuccessfully waiting 10 secs for the quad to descend, (RTL_LOIT_TIME set to 2000ms) I switched to Loiter and landed myself (@line 156000).

Any ideas what caused this?

Hi Mark, I had a look at this. This one is an interesting situation…

In the CTUN message, there is no DCRt registered during the RTL events. Maybe this is normal, I’m not sure yet.

But also, during the second RTL event, the copter does not seem to be hitting it’s target altitude very well. Like it’s just not trying hard enough. You also have some divergence between the CTUN/BaroAlt and the GPS/RelAlt. This is often caused by vibration. Your vibration is actually fairly good… when you’re not moving. But quite bad when you’re flying at speed.


This is actually not uncommon, but it’s not helping the situation for sure. Does this machine have large props?

I’m not sure yet on the descent rate control though. You say that it should slow down to Land_Speed at 10m. I’m not sure how that part is done, I’ll have to look into that. But FWIW, the system thought it was still over 11m when you switched to Stabilize. Even the Baro Alt is just slightly above 10m. You’re saying it was much lower than that?

Hi Rob, thanks for taking a look at this.

The copter is running a 4S battery and has 9" Graupner E-Props on 1000kV motors, a fairly responsive combination. You’re right about the vibrations, I’m not sure why they were so high back then - they seem to be improved now though.

You’re right, I switched into stabilize when the logged altitude is at 13.3 m - that explains why it didn’t slow down.

From the video (I just added some annotations to clarify), you can see it doesn’t slow down and actually seems to speed up it’s descent.

The logs don’t show this, CRATE actually indicates it did slow down it’s descent, and that it only sped up when I switched to stabilize, but this doesn’t reflect my memory of the event.

I got a scare when I noticed it suddenly speeding up, this made me quickly switch to stabilize.

Regardless, it’s possible it might have slowed it’s descent once the sensors detected 10m, despite the strange speedup. I’m happy with the explanation that this was caused by excessive vibrations, causing an inaccurate altitude reading. Haven’t noticed anything similar since - auto-landings have been working perfectly every time.

This is a vibration issue. Look at EKF4.SH and EKF4.SV and EKF3.IVD (documented here: … view/#EKF4)

Thanks for that Jonathan.