Spikes in desired pitch on fast, downhill waypoint mission

I have been flying autonomous waypoint missions down a large hill, and am interested in getting the quad to fly as fast as possible while maintaining nice, smooth flight characteristics. As I was increasing the horizontal and vertical acceleration and speed parameters, I began to notice a discernible twitch in the quad’s flight as it descended the hill. Upon inspecting the logs, it appears that each twitch coincided with reaching a waypoint, and the desired pitch seems to spike and the desired altitude flattens out. I was hoping someone could help me understand this behavior, and would welcome any suggestions for how I might be able to smooth my flight out. I am running Arducopter V3.2-rc8 on an x8 equipped with a Pixhawk. Also, here are some parameters that I tweaked.

wpnav_accel: 400
wpnav_accel_z: 250
wpnav_speed: 1500
wpnav_speed_dn: 500
wpnav_speed_up: 750
inav_tc_z: 7

I apologize for providing the .log file rather than .bin. Unfortunately I had already converted and cleared the logs. Also, as evident from the logs I had a pretty bad crash towards the end of this flight which I believe is a result of the changing air pressure around the quad, which resulted in the barometer thinking the quad was higher than it was which caused it to run into the ground.

Thanks for the help!
-Gabe

Are you using spline waypoints?

No they are not

Try using spline waypoints - they may be smoother.

Baro dynamic pressure problems are really hard to solve. We need a baro probe designed to solve them, and 3DR doesn’t seem to be interested in sticking a little hose barb on the pixhawk housing… maybe they’ll make us an external baro at some point. It’d be really nice if this problem was solved.

By the way, I will make sure this gets to the appropriate people. I am of the impression that 3.2 isn’t supposed to twitch at waypoints.

Sounds like this won’t be getting fixed in 3.2. We do have an idea of why it is happening though.

I may take a personal interest.

Spline waypoints seem to work well, thanks Jonathan!

Ok, good information, thanks.

By the way, I’ve recently been doing experiments with barometer things. Drops in altitude are due to low pressure - which makes it think it is too high. This primarily happens for most people when they lean back to decelerate after getting the quad up to speed. Is this what happened to you?

If so, it can be alleviated by increasing EKF_ALT_NOISE to 2 or 3. This will substantially lessen the weighting on the barometric altimeter and instead trust the GPS vertical velocity and the inertial system. It will make the altitude more sensitive to GPS glitches and vibration and the like, however.