Servers by jDrones

Dev Call July 8 2019 2300 UTC

(Craig Elder) #1

GSOC

  • Student Updates

Issues & Pull Requests


















Plane

Copter


Rover

(Craig Elder) #2

Attendee count (max): 19

UTC2301 - GSoC update

  • P Hall
    • Kalman Filter stuff is working and matrix/template stuff worked out
      • Outputs look sane
      • Working in a mathematical sense
      • Needs tuning for flight
        • Lift coefficients
        • Gets an answer but it is hard to say if we have the right answer
      • PR coming
      • Blog post coming
      • Next step is a prediction of endurance of the aircraft
        • Controlled simulation system should be able to give you the relevant expected output?
          • Training data is the same as test data?
            • Use different flight speeds to mix things up
        • Turn off unlimited battery!
      • On sailboats it would be nice to use this for navigation
  • Akshath and MAVProxy
    • Gtk / Wx issues under Linux causing issues
    • Great progress!

UTC2314 - https://github.com/ArduPilot/ardupilot/pull/11742

  • Need to check Michael’s issue

UTC2319 - https://github.com/ArduPilot/ardupilot/pull/11735

  • Need to make sure this fixes the issue it is claimed to

UTC2323 - https://github.com/ArduPilot/ardupilot/pull/11725

  • Original author hasn’t gotten back
  • merged

UTC2325 - https://github.com/ArduPilot/ardupilot/pull/11722

  • Merged!

UTC2326 - https://github.com/ArduPilot/ardupilot/pull/11672

  • Unrelated: Calling singletons from interrupt handlers is bad if the singleton can do memory allocation
    • This is not our singleton pattern!
    • Generally true of memory allocations in interrupt handlers
    • Caught by using the --enable-asserts option
      • ./waf configure --board xxx --debug --enable-asserts
    • Could add a sanity check that we don’t call memory allocation in new and malloc
  • Header cleanup will be done as another PR

UTC2332 - https://github.com/ArduPilot/ardupilot/pull/11671

  • Object avoidance
  • Testing showed load was minimal
  • Array-new and array-delete
  • Need to make sure new is deleted and malloc is free’d

UTC2334 - https://github.com/ArduPilot/ardupilot/pull/11669

  • Will check binaries are the same

UTC2341 - https://github.com/ArduPilot/ardupilot/pull/11623

  • Default of 200?
  • Tridge is dubious on having a parameter
  • Transmitters can debounce switches
  • Tridge thinks he might have a preference for the counter
  • Moving switches between positions
  • 1/3rd of a second too long?
    • Rob says he lost a vehicle because of this 1/3rd of a second
  • Resolution: no parameter, just a #define

UTC2348 - https://github.com/ArduPilot/ardupilot/pull/11482

  • Read-only by name?
  • Hard-limit on distance travelled (250km) is South Africa
  • Apj-tool to append a set of read-only parameters by name?
  • Will defer until after call tomorrow
  • Indian legislation
    • Extreme case to protect
      • Same ability to update firmware as manufacturer has
      • So a fixed firmware
        • To meet GPL requirements
        • Cube with certified version of firmware
          • Would have to swap the Cubes to update firmware
          • Remove USB connector enough?
            • Would need a jtag connector to update
  • Hot-patching code to remove restrictions is doable
    • E.g. DJI with stayout zones
      • With a ublox?
        • Hack hardware so that attiny between ublox and main flight controller to change the apparent coordinates the FC sees
  • Always possible for end-user to bypass
    • Replace stm32 if required!
  • Not enough information on the legislation in any of the countries to work this out
  • “What is enough?”
  • Rob thinks “best effort” is enough

UTC2358 - https://github.com/ArduPilot/ardupilot/pull/10433

  • Needs Randy
  • Treat guided-nogps like guided

UTC2359 - https://github.com/ArduPilot/ardupilot/pull/10178

  • Allocating two baros would be nice into the future
  • Merged

UTC0001 - https://github.com/ArduPilot/ardupilot/pull/7215

  • Tom added devcall
  • No Tom
  • Hold-up is Paul wants a different API inside EKF used for it
  • Redux 0008 due to Tom returning
    • Needs rebase
    • Needs to address Paul’s comments
    • Can be tested in SITL
    • Should be tested on real hardware
    • This PR isn’t for dual-GPS
    • Some GPS types can provide yaw with a single GPS
      • E.g. Hemisphere GPS
        • Good for Rovers
      • E.g. some Septentrinos
    • Does not do multi-gps yaw
      • But is a precursor

UTC0002 - https://github.com/ArduPilot/ardupilot/pull/11720

  • Going into stable betas
  • Don’t usually put changes like this into stable
  • Issue in April with failing IMUs has not been fully addressed before these patches
    • Believed to be fully addressed as it can be afterwards
  • Doesn’t work on current accel or rate, the work on delta-angles and delta-ts
  • When IMU comes back it could have an accumulated delta-time and delta-accels much larger than normal (as in 3 orders of magnitude)
    • EKF’s attitude would change by 90 degrees
  • Flown by several people
  • We were losing IMU samples frequently under high load
    • Smoother flights under heavy load

UTC0008 - https://github.com/ArduPilot/ardupilot/pull/11676

  • Flying absolute altitude missions? Definitely want to run these patches
    • Could end up with >100m altitude error
  • Continuous height-datum while disarmed issue
  • Nasty bug
  • Related to EKF origin drifting
    • To poor values
    • 120m of height drift has been seen

UTC0015 - Plane update

  • Betas went out yesterday
  • Flight logs and results would be good!
  • Please test!
  • Plane 3.9.9 release
  • EKF using earth’s magnetic field model?
    • Discussed previously
    • 6 numbers to learn
    • Earth’s magnetic field and offsets
    • Declination was used for the earth’s magnetic field
      • It sometimes diverged and it stopped using the compass
      • Hovering without compass for long enough you will lose yaw and get a fly-away
      • Planes can recover using fly-forward assumption
      • Lots of logs showing it diverging
      • We now constrain divergence using world-magnetic model
      • Can learn offsets in-flight much more quickly, too!
    • Lying about your location (e.g. with a mavlink GPS) will be a very bad idea

UTC0020 - close