DEV call 20 Feb 2017 3pm PST = 20 Feb 2017 2300 UTC

Canberra Meetup

13 new photos added to shared album


  • 3.5 update


New Mixer which leads to
Better mixers for Arduplane, PX4, mavlink etc
deep stall:
how to handle LAND_DS_ENABLE to disable sub-groups of params




  • BBBlue
  • CANESC First flight

0:02am - tridge and roundup of the dev meetup
Randy was here early, so a lot of stuff happened on Thursday
FOC control ESC demonstrated in-flight on a quadcopter
Vast number of people there
Many only leaving Canberra now (Tuesday morning)
Total fire ban on Saturday
Hovered 3D plane briefly in pool area at hotel
Brief hover in Randy’s room at the hotel
Flew 3D plane in small park, full hover
No boats or submarines
Next time we’ll make use of a lake
RTK gps’s
3D plane
Hge success
hover/loiter/transitions worked like in simulator
APM2-based quadcopter
Copter 2.9, laser-cut frame, mtk GPS, …
Big trad-heli
Follow-me’s between various vehicles
LEDs on a whole bunch of vehicles
Inverted landing of Firstar
Talks on
Lots of developers
Have been recorded, Randy will upload
Commitments in front of whiteboard
Blogs will be coming
All want to do it again
More notice next time so more people can come along!
Randy will be scarce for a few days as he’s travelling

10:10am - 3.5rc1 update with Francisco
Randy planning on rc2
Not much to report on rc1
Everything is going really well

10:12am - PX4 mixing in realtime over MAVLink
Hoping to get that dev to help with new-style mixing
Tridge points out that we only use the IO mixer in override condition
It basically doesn’t happen
You can set it up as an override condition etc
Very unusual
We want the new mixer to be active all the time
Tridge is short on time, hoping Matt can find time to do it as a “new-style” thing
Current PR is very unlikely to go in as we move away from the px4-style IO mixer
Wouldn’t work on Linux ports for e.g.
More ties into PX4 which we don’t want ATM
Tridge understands motivation, but it’s not the correct path for ArduPilot

10:16am - deepstall
Hiding deepstall parameters with the hide
Land_type and land_deepstall_enable
We now can probably do without the deepstall_enable and rely on the land_type with tridge’s most recent commits for parameters
Need to find a syntax for more complicated parameter-showing
Having a bit-per-set of parameters is cheap for storage but we might run out of bits
Would be nice to have a function pointer instead, but that’s 4-bytes per group-info (400-500, perhaps), which would be 2k of flash.
Might be able to have a groupinfo_conditional() macro and burn an entry in the table to store a value which store something like “GROUP_CONDITIONAL(type, 4))” which would hide further entries unless type is 4
Might add sub-tables without having a separate class
Almost ready to go!

10:09am - callback for onChange() on parameter
Rejecting parameter changes?
E.g. changing landing type in middle of a landing
But there are vast numbers of things you shouldn’t changing in-air
Too much flash being burnt on protecting users from being complete idiots
Read-only / only-when-disarmed would probably do the job
onChange might be able to eliminate shadow-copies of variables
Can probably just go off the object address for this
Linked list, probably
Any parameter change would scan the linked list
So the CPU argument is dubious

10:46am - travis dying because of docs issues
MdB found the bug - live!

10:48am - soaring PR

Someone has resurrected an ancient PR and rebased it on 3.7 and created a new PR
Original author popped up (ST)
He’s happy to see his stuff go in!
Sparse vs dense matrices
We tend to use matlab to expand stuff out
More code
But less data
4-state matrix
Tridge is happy with the dense approach
It is more CPU intensive
But much larger flash footprint to go sparse
100kB for the 24-state EKF
With a 4-state should be manageable as a dense matrix
Runs on APM2.5, sooo…
Matrixn.cpp is where?
Nowhere ATM, but we should have one
Probably dynamically allocated
Be wary of dropping one of these on the stack
Do the allocations at startup if soaring is enabled
This can be done later as we’re not currently short on RAM
ST is happy to rejig things
MatrixMath.cpp should move to using stdint types
For consistency across platforms
#pragma once in headers
Scheduler table fixes required (change in units hertz vs decimator of loop rate)
Tridge meant to get to the old soaring stuff
Was really happy to see Andre had picked it up
This PR isn’t quite ready to go in, put not a long way off
Tridge wants to look at some of the logic changes in Plane
Need to look at flash space required
Double-precision maths isn’t a good thing (exp and pow rather than expf and powf)
PX4 strikes again
Docco required in Wiki!

11:04am - Randy pops up
In avocado farm
Quick Copter update for 3.5rc
Rc2 due out within a week
Patch from Paul for GPS blending might go in
EKF failsafe issue resolved
Changing declination not a valid way to test
Fixes for new LED display
MdB’s uBlox patch
Randy will rebase on master
Vision stuff
Zed-3D camera odometry will be integrated into EKF
Doesn’t use ROS
Z-Camera comes with algorithms to do things like odometry
Currently fake GPS messages
Has been flown
Zed-Cam has no IMU
Must be forward and level
There’s another partner company working on other vision stuff using ROS
AT and crew
Position estimation not done yet
TP brings up RealSense
Zed would seem to have longer range
Maybe 15m

11:15am - Tom and loiter
Loitered forever
MdB’s recent altitude fixes might resolve this one
Trouble getting log because of translation issues
If this one’s not sorted it is a blocker for 3.8

11:22am - ublox change with MdB
We stopped sending initial config string to avoid problems on m*p
Broke badly-configured uBlox devices
New patch is up, should be put in next plane and copter

11:24am - ArduSub merge
This PR looks like it is ready to go in!
Tridge is happy, tridge is happy, Randy is flying quadcopters right now
Tridge will work on processes with Jacob about how to keep this updated
QGC is recommended as ArduSub uses the MANUAL_CONTROL messages and nothing else supports it ATM
BB might want to add support to APMplanner2
Make sure SITL stuff works
Wiki needs to be updated to reference sub
Announcement on
Really, really good PR - docs and rework and everything

11:34am - Philip and hardware
Canbus escs flown!
Jonathon just wrote it all, did everything from scratch
About 6 months
Sensorless and sensored
Sensored allowed validation of EKF
Really good timing to be able to show enRoute in Canberra
Lots of CANBUS happening soon
PR will be sponsoring JC to do more CANBUS
More sponsorship would be appreciated….
F3 chips
Later will be based on something else
But more important to get stuff out there
Based on old PXF work + BeagleBone Black
28mmx28mm chip is a BeagleBone Black!
Single board has WiFi, BBBlack and the bits from the PXF
Supported in ArduPilot now
Really good demonstration of how Open Source hardware should work!
IMU is 9250 on i2c
Should have been SPI
That i2c bus is going to be very, very busy
Too many sensors
Maybe not good for Copter
If IRQ line from MPU9250 is connected might be able to use it
No ethernet
New plug?!
Another opportunity there for a revision…
Really cool board
Starting to work out how to put together this redundant commercial autopilot system
Redundant all the way out to the ESCs
5 CPUs
Not sure if it will be open-source hardware
Schematics will be open

11:55am - off the floor
Tridge found and fixed a bug in the servo libs which he’s pushed
TP mentions
Still currently with uAvionix but someone should look at it
FF will look at it
PR Yona in Singapore
Founder Tian
Big heavy-lift copter they want to design for flying in PNG
Chinook-style heli
Need to put in a petrol chinook into the sim
There’s a patch out there to add support for Chinook-style vehicles
Should fit in better now
Ospreys coming!
Vertical accuracy from GPS
User log showed issues
Patch looks good

1 Like

APM Planner 2 already supports manual control message. Not sure I’d use it in preference to QGC. APM Planner 2 graph feature is really its shiny thing! Check it out.