Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Andrew Tridgell
    @tridge
    for that one we set the circle radius to 80cm. We need a bigger room :-)
    Andrew Tridgell
    @tridge
    @anbello I've done a PR for a wiki page on Vicon setup here: ArduPilot/ardupilot_wiki#1983
    I'm working with Hon from ANU on doing aerobatics in the Vicon room. We're starting by trying to get it tuned so it can do CIRCLE mode as fast as possible. After that we will try to get tight flips
    then more complex flight paths
    Andrea Belloni
    @anbello
    Thanks @tridge really interesting
    Andrea Belloni
    @anbello
    @tridge after reading new wiki page some questions:
    How can you use GLOBAL_VISION_POSITION_ESTIMATE without this PR (ArduPilot/ardupilot#10374) ?
    Why GLOBAL_VISION_POSITION_ESTIMATE instead of VISION_POSITION_ESTIMATE in indoor?
    Why modify GLOBAL_VISION_POSITION_ESTIMATE message for velocity instead of using VISION_SPEED_ESTIMATE as in this PR (ArduPilot/ardupilot#9505) ?
    Andrew Tridgell
    @tridge
    @anbello we can use the vision msgs with EKF2 now. The PR #10374 is needed to use it with EKF3. Right now we can either use GPS_INPUT with EKF3 (as EKF3 supports yaw input from GPS) or use vision msgs with EKF2. It would certainly be nice to support both methods with both EKFs
    At the moment GLOBAL_VISION_POSITION_ESTIMATE and VISION_POSITION_ESTIMATE map to the same thing in ArduPilot. The reason I like using GLOBAL_* is I like to be able to look at logs and GCS and treat the GPS data in the log as truth data. I know we have the VISP msg in the log, but there are better tools for displaying GPS data. As GPS data comes directly from the Vicon when using GPS_INPUT it is actually a pretty convenient way to work
    The reason I didn't use VISION_SPEED_ESTIMATE is that it comes as two packets, one for position, one for velocity. The way the posvel fusion in both EKF2 and EKF3 is structured at the moment it assumes that the two sources of data come in with the same timestamp. Doing it as two packets breaks that assumption. I haven't looked at #9505 enough to know if that resolves the issue
    Andrea Belloni
    @anbello
    Thanks for exhaustive answers @tridge
    Andrea Belloni
    @anbello
    @tridge looking at mavproxy_vicon.py I see you send GPS input messages at 5Hz and vision estimate messages at 14Hz, lower rate external navigation control should get less precise control with complex trajectory, maybe I am wrong, what you think about It?
    Andrew Tridgell
    @tridge
    @anbello yes, I used 5Hz as that is default for GPS, but I plan on testing higher rate GPS data next Friday. I'm only in at the ANU vicon lab once a week. EKF2 has a limit of 14Hz (actually 70ms period) for external vision data: https://github.com/ardupilot/ardupilot/blob/master/libraries/AP_NavEKF2/AP_NavEKF2_Measurements.cpp#L888
    I don't see an equivalent limit for GPS data in EKF3, but I'd like to test higher rate data before putting it in the module and wiki
    if you wanted to test higher rates for GPS_INPUT and confirm it works I could change it before next friday
    I guess we could test it in SITL ....
    the other thing I'd like to do is properly measure the lag. The GPS_DELAY_MS=50 is just an educated guess
    and at higher accelerations that delay will become really important
    this is what it looked like when doing circle at 120 degrees/sec:
    Andrew Tridgell
    @tridge
    image.png
    i think the offset is because we didn't place the copter at 0,0 when it started, so EK3 origin was not the same as VISP origin
    here it is with GPS data:
    image.png
    pushing GPS rate to 20Hz or so would certainly make for nicer graphs if nothing else :-)
    Andrew Tridgell
    @tridge
    i'm looking fwd to next Friday as we plan on trying to keep pushing circle rate up till it breaks. I don't know what the limit is, but I do hope it is over 360 deg/sec. I think it would look cool twirling around that quickly
    Andrew Tridgell
    @tridge
    this is what it looked like on the map at 120 deg/sec:
    image.png
    patrickpoirier51
    @patrickpoirier51
    @tridge circles are not all made equal ..;-)
    image.png
    50 cm CIRCLE at 50 Degs/Sec with T265 looking downward 2 Meter altitude
    patrickpoirier51
    @patrickpoirier51
    Sometimes it makes a pattern that looks like Epitrochoid curves
    image.png
    Andrew Tridgell
    @tridge
    @patrickpoirier51 pretty :-)
    Shervin Emami
    @shervinemami
    @patrickpoirier51 you made an automated basket weaving drone ;-) But seriously, that's still quite impressive precision for such a small & fast circle!
    patrickpoirier51
    @patrickpoirier51
    😂
    guglie
    @guglie
    @tridge very cool, I'm trying to do similar tuning for GUIDED pos control, by the way how do you plan to measure the GPS_DELAY? For now I've moved the drone back and forth and then visually compared accelerometer data with position data with a small python tool I made.
    patrickpoirier51
    @patrickpoirier51
    Some more tests with the T265 on a 330 Quad
    image.png
    Still working on the scaling error (setted @ 1.8 for the moment)
    image.png
    patrickpoirier51
    @patrickpoirier51
    image.png
    image.png
    As you can see it is tracking using these white corrugated sheets with black tape patterns (random shapes)... and my SUV ;-)
    Andrew Tridgell
    @tridge
    @guglie for the tests I'm doing yaw is available from GPS_INPUT, so I'm planning on looking at gyro vs GPS.Yaw on a sharp yaw
    guglie
    @guglie
    @tridge Thanks, unfortunately I have no yaw from positioning system.
    guglie
    @guglie
    What mode are you using? I'm using GUIDED. In the following post I've had some support from Leonard, maybe it can be useful to you, but he said that he didn't write the control for GUIDED so he can't help for the issues in that part.
    https://discuss.ardupilot.org/t/position-controller-pids-missing-pos-i-inconsistencies/44582/8
    patrickpoirier51
    @patrickpoirier51
    @rmackay9 24 ghz radar for avoidance.