Extensive drift on RTL landing sequence on RC4

Was testing out 3.4 RC4 on the hexa, and twice during an RTL mode the drone would return to hover over home but as soon as it began the vertical descent portion of the mode it would start drifting, not unlike it can in Land mode in wind conditions. Also (and both of these might all be red herrings) it is not too different from what I’ve seen when the Pixhawk’s computing resources are overtaxed.

The drift was many meters over the course of the landing, and therefore extremely unlikely that this is proper behavior. System otherwise flew stable and well in Auto and Loiter and had good GPS lock the entire time. This RTL landing behavior is different from what is seen on vanilla Arducopter 3.3.1. On 3.3.1 it generally did what you would expect, i.e. fly to home position and land on it.

More details:
RTL was run three times, and this behavior was observed on the first and third time. First two times it was used the drone went from Loiter mode, second time it was from Auto.

All sensors (accels, compass, RC, leveling) were carefully calibrated and the system power cycled immediately prior to this flight. Wind was relatively slight and previously would cause negligently small errors on RTL.

Any thoughts on root cause?

DataFlash logs failed to attach directly to post perhaps because of large size. Here are the links to the same log in two different formats on my dropbox (.LOG is compressed to smaller size for slower connections).

https://www.dropbox.com/s/n1zowlxdnw97rhw/53.BIN?dl=0 (.BIN, 205MB)
https://www.dropbox.com/s/j2yx8lw91rnr03y/53.zip?dl=0 (.LOG in .zip 97MB)


Is it possible that the tabs on the transmitter are not in the center? It looks a bit like this during the RTL that the roll and pitch sticks are down around 1400 but the RC range is about 1100 ~ 1900. Maybe an RC calibration needs to be done if the transmitters been changed?

The vehicle looks generally under control.

Interesting. Will retest and report back. Thanks for the insight!