Servers by jDrones

Dev Call July 1 2019 2300 UTC

(Craig Elder) #1

Copter**

GSOC
Student Updates

Issues & Pull Requests








Plane
Update

Rover

(Craig Elder) #2

Attendee count (max): 18

UTC2301 - Randy and Copter and Rover

  • Still working on OA ahead of Copter 3.7 and Rover 3.6

UTC2302 - https://github.com/ArduPilot/ardupilot/pull/11671

  • Object Avoidance
  • Some issues on OA
  • Tom’s issue and some smaller ones
  • Testing today on hardware
  • Works very well for Tom
    • Getting cool maps in MAVProxy in SITL with ROS-like maps being generated
  • MAVProxy PR has been updated
  • A lot has been removed
    • More basic, more stable
    • Either initialises or permanently disables itself
    • Creates DB
    • Whenever you see something it populates the database with earth-frame locations
  • Tridge would like an options bit to display on the ground-control-station via adsb messages
    • A debug option
    • 800 to 1500 ADSB messages/second in debug!
  • Allocation failure isn’t being checked correctly?
    • There’s a non-obvious thing checking it indirectly
    • Doesn’t free first allocation if second fails
    • We allow free of nullptr
  • The ADSB stuff is quite hacky
    • Throttling is done within database library
    • Randy didn’t want it in as he wants the stuff merged
  • Doesn’t seem to check if we’re on a 1MB board
    • Needs to check for minimise-features
    • New features shouldn’t go in the 1MB boards
  • Some of MdB’s issues have been addressed
  • Update does a 10Hz update/optimisation then a queue draining as fast as it can
    • Signalling would be good
  • Maxing out CPU and maxing out RAM
    • Is it practical?
    • Rover has lots of spare CPU
    • Will it actually work on Copter?
    • Currently running on Cube on Rover
    • RAM seems to be current limitation
      • Default config is 50 elements
      • Unknown RAM consumption
      • Location and timestamp
  • What size is an object?
    • Item from proximity library is a point
    • Within filter radius of a point means you get filtered out
    • Old point gets timestamp updated
    • More of a point-cloud database
    • Objects don’t have sizes
      • A wall will give you lots of “objects”
      • If there’s also a pole you might lose the gap between the pole and the wall
    • Optimised for capturing data not for storing it
    • Are we capturing the stuff we want?
  • This is functionally working, if it’s working it should go in (
    • If we’ll have to restructure it should we actually merge it?
    • We’re missing protections
    • Does this make sense long-term for where we want to go?
    • We can change the structure later
    • Too hard to say if this will work on a real vehicle
    • We should write down all the things Tom should do
    • We should compare with what we currently have
      • 8 points!
        • We forget them
        • We don’t avoid them
    • Tom is afraid of year-long feature-creep
    • How much done on hardware?
      • Still being tested
      • Tom is working on realsense direct drivers to make things better
      • 8 sectors + large database
  • Scaling logic unlikely to be final form
  • Separate plane avoidance would be required
    • Not storing size
    • Not storing velocity
  • This stores proximity data in earth frame
  • Currently only fed from proximity library
  • Scanning lidars can give us sectors of gaps
    • Sf-40c
  • The OBC2018 code already had a database which did expiration etc etc
    • No proximity sensors in Plane
  • This could be made to work for the OBC case
  • 9:33 AM] (Channel) davidbuzz: a dense straight line of circles does a reasonable approximation of a boundary and the distance/s we want to stay away from it.
  • [9:34 AM] (Channel) davidbuzz: I think it would be pretty easy to add velocities and moving points in the future for OBC.
  • [9:37 AM] (Channel) davidbuzz: I don’t think this and OBC need to be the same thing right now. keep them separate and let them evolve separately, and maybe one will absorbe the other at
  • MUSTS include HAL_MINIMISE_FEATURES

UTC2338 - GSoC student updates

  • Peter Hall
    • KF stuff nailed down
      • lift/drag/thrust/power
      • May be able to answer “how far can I go?”
        • Work out AoA/drag/thrust and look at battery remaining
        • Rectangular matrix stuff required
        • 11x3 rectangular matrix
        • 11x11 multiplies
        • Need to check CPU cost
          • Different thread required?
          • Don’t forget to disable interrupts while measuring
  • Akshath has been making good progress on Parameter module
    • Currently throwing annoying messages from gtk library
    • Tells users not developers
      • Akshath doesn’t see warnings….
  • Rajat and Airsim
    • Lockstep wasn’t working with old apporach
    • Lockstep is working in new approach
      • Low innovations
      • Framerate is very low
      • Coordinate cleanups will help a little
      • Chasing realtime comes after that
  • Matt and tradheli
    • Working on entry and glide
      • Controller of rotor rpm using collective
      • Trying to use a PID to control rotor speed
      • Using a low-pass filter on the collective position so that it can get a trim; lpf would follow trim of collective while pid is working so it wouldn’t require a large interim offset in the collective
    • Almost there!
  • Patrick update for t65 work
    • [9:47 AM] (Channel) patrickpoirier: (microphone not working) We started Beta tests : 3 testers for the moment, we need to optimize the T265 initialization sequence so we can get good localisation in bothh forward looking and downward looking.

UTC2343 - https://github.com/ArduPilot/ardupilot/pull/11681

  • Reduce usage of inertial nav
  • Supposed to be mostly non-functional
    • Functional change has been commented
      • Seems to be a sensible change
  • Using set_alt_cm OK?

UTC2353 - https://github.com/ArduPilot/ardupilot/pull/8364

  • Remove mtk binary drivers?
  • MdB broke this a long time ago and only NightGhost noticed
    • 2017
    • MTK don’t persist configuration?
  • MTK binary protocol doesn’t give you vertical velocity anyway
  • Resolution one way or the other?
  • These are very-little used
  • Needs a rebase

UTC2358 - https://github.com/ArduPilot/ardupilot/pull/11691

  • Fixes F9 autoconfiguration
  • Other pre-existing bugs won’t be fixed as part of this PR
  • Variable to be removed
  • Backports will be done so community start to move towards F9
  • Testing with CSGShop F9 as it fits in a vehicle

UTC2359 - https://github.com/ArduPilot/ardupilot/pull/11676

  • Consumed a ridiculous amount of time
  • Made a lot of progress on a call this-morning
  • Behaviour change will happen
  • New stable release will be required
  • Flying absolute-altitude may lead to diving into ground
    • Absolute altitude can diverge by unlimited amount
    • Impacts all vehicles but Plane mostly
    • Tridge will continue work on the PR
    • Gateway patch for new stable Plane release

UTC0002 - https://github.com/ArduPilot/ardupilot/pull/11664

  • Merged

UTC0003 - https://github.com/ArduPilot/ardupilot/pull/11663

  • Merged

UTC0003 - https://github.com/ArduPilot/ardupilot/pull/11652

  • Merged

UTC0005 - https://github.com/ArduPilot/ardupilot/pull/11596

  • Has to be done in a different fashion
    • Will break things this way
  • Probably sending generic and rely on GCSs to update their interface when a more-specific type is received
  • Do we need an “unknown” instead of just “generic?

UTC0009 - https://github.com/ArduPilot/ardupilot/pull/11698

  • Needs a rebase
  • Can be merged once it’s ready

UTC0012 - https://github.com/ArduPilot/ardupilot/pull/10793

  • Flies a v22 osprey
  • merged!

UTC0013 - Plane update

  • Tridge is working hard on altitude thing
  • And a couple of integrator-related things

UTC0017 - Jani is home

  • Working on videos from Shenzhen