Dual-motor tailsitters

tridge · April 10, 2017, 5:37am

this has been fixed now

kd0aij · April 10, 2017, 2:24pm

@tridge I flight tested the stabilize to hover transition on my last flight (the one that ended with a burned-out motor) and it was smooth: https://drive.google.com/open?id=0Bw3digSMQXDuUlYyR092OThPU00

tilt · April 10, 2017, 7:42pm

1/4" 6mm+

in shop, even better?is this what you want?

tridge · April 10, 2017, 8:12pm

the additional mass looks great. Any reason why you have 3 cubes of the zeal instead of 4? I could imaging that may oscillate a bit with 3? You could judge it a lot better than me though, so if it feels OK then ignore me.

tridge · April 10, 2017, 8:13pm

@tilt also, is that steel wool under the flight board?

tilt · April 10, 2017, 8:16pm

no foam,air filter [quote=“tridge, post:210, topic:15302”]
Any reason why you have 3 cubes of the zeal instead of 4?
[/quote]

there are 4 just opposit corners keeps me away from the green goo

tridge · April 10, 2017, 9:20pm

ahh, makes sense.
I’m really looking forward to you trying the new code with thrust vectoring support!

tilt · April 10, 2017, 9:37pm

me to! the system says I talk to much.

tridge · April 10, 2017, 10:01pm

it tells me that too, I ignore it

tilt · April 10, 2017, 10:11pm

OK, want pic of motor mt, for friends build

tilt · April 11, 2017, 6:33pm

Hi, Tridge
starting to work the control system, tilt must be active when not armed, props will dig holes
only getting 20 deg tilt how to add tilt deg.w/o elevons? tilt servos are set to 180 deg.
mixing is all messed up, should I sort wbwa 1st then hover 2nd than manual 3rd ?

tilt · April 11, 2017, 8:58pm

sorted controls

kd0aij · April 11, 2017, 11:43pm

@tridge Hover tuning the Stryker with the new motors: was initially unstable in qstabilize on roll axis, then encountered oscillations due to coupling between pitch and yaw. Logfile: https://drive.google.com/open?id=0Bw3digSMQXDuMU8yc192VVdwZTg

Also, flight mode background color seems to be wrong in MAVExplorer; qstabilize is labeled fbwa

tridge · April 12, 2017, 1:26am

I notice you didn’t have GPS lock. At the moment that means it will be using DCM, not EKF. I think we need to change that. That log is mostly FBWA mode, with just 12 seconds of QHOVER, and I see large discrepencies between attitude estimation between EKF and DCM for that time. I don’t think you necessarily have a tuning problem, I think the most obvious issue is in state estimation.
There are probably tuning issues too, but if attitude estimation is off you can’t tune.

tridge · April 12, 2017, 1:27am

Just a note to everyone that I’ve pushed the changes for vectored thrust support to master. These changes also mean that the ESC outputs for tailsitters now run at Q_RC_SPEED (default 400Hz). They were at 50Hz previously.

kd0aij · April 12, 2017, 2:28am

@tridge That last flight was on latest master, so it should have had 400Hz ESC outputs, and mavproxy claimed I was in qstabilize and qhover mode, not fbwa. Is is it possible to hover in FBWA mode? The motors were spinning when armed, which is not the case in FBWA. I did at least 6 takeoffs and upright landings in that log.

I"ll have to pay closer attention to system status. I don’t know why there was no GPS lock; maybe the cable is damaged. Will have to get the pre-arm checks working properly; do they require just a 3D fix, or are there also constraints on hdop/vdop as in PX4?

tridge · April 12, 2017, 2:36am

there are quite a few constraints, and not just on hdop. See this function:

github.com

ArduPilot/ardupilot/blob/master/libraries/AP_NavEKF3/AP_NavEKF3_VehicleStatus.cpp#L25


      
          if (inFlight && assume_zero_sideslip() && !use_compass()) {
              // this is a special case where a plane has launched without magnetometer
              // is now in the air and needs to align yaw to the GPS and start navigating as soon as possible
              gpsGoodToAlign = true;
              return;
          }
          
          // User defined multiplier to be applied to check thresholds
          ftype checkScaler = 0.01f*(ftype)frontend->_gpsCheckScaler;
          
          if (gpsGoodToAlign) {
              /*
                if we have already passed GPS alignment checks then raise
                the check threshold so that we have some hysterisis and
                don't continuously change from able to arm to not able to
                arm
               */
              checkScaler *= 1.3f;
          }
          
          // If we have good magnetometer consistency and bad innovations for longer than 5 seconds then we reset heading and field states

lorbass · April 12, 2017, 1:30pm

First Takeoff ended as Flip vorward. (No transition in low level:slight_smile:)
After analysing Log and video I think it was my fault.
Due to verry slow (carefully) increasing throttle the Tailsitter tilted forward, the elevons couldn’t not correct it, because the landing gear out of the winglet (2mm carbon) where stuck to the ground.(good to see in the pic)
And then immediatly increasing the throttle was to late probably not enoughe power to lift with this angle.
Video: https://www.youtube.com/watch?v=-l7qmgBgiQc&feature=em-upload_owner
Frame OK, Props damaged.
So, the conclusion: Full power for the takeoff.
Or other advice?

kd0aij · April 12, 2017, 1:35pm

@lorbass Yes, takeoff is a bit difficult with this kind of tailsitter. I throttle up slowly till I’m sure the motors will respond quickly, then do a jump takeoff by (nearly) firewalling the throttle. You definitely want to clear the ground quickly with a lot of thrust for elevon control. Also don’t back off the throttle too far after you get airborn, try to avoid quick descents until it’s tuned well.

I did something similar with the “landing gear”, but used pivoting legs to avoid breakage during horizontal landings: https://goo.gl/photos/aLYKJQJtRoqs9sGs9

kd0aij · April 12, 2017, 1:38pm

@tridge I think my mode changes weren’t getting logged reliably. Seems there are a lot of dropouts around mode changes: https://drive.google.com/open?id=0Bw3digSMQXDuaGVBZkVReklpNjQ
I always arm in QSTABILIZE mode, then switch to FBWA to disarm. The 10 second disarm delay is quite inconvenient during testing.