Dev Call Jun 8, 2026

CraigElder · June 8, 2026, 10:58pm

CI Workflows

Release Update

github.com/ArduPilot/ardupilot

ArduPilot 4.7 issues list

opened 12:30AM - 07 Mar 26 UTC

rmackay9

DevCallTopic ReleaseAdmin

This is a list of issues discovered during ArduPilot 4.7 testing. [Issues for 4….6 are here](https://github.com/ArduPilot/ardupilot/issues/28612) ### Reports requiring investigation - [ ] still GCC 10? ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta6-is-available-for-beta-testing/144106/3)) - [ ] Rover jump after 8+ hours with RTK GPS ([discussion](https://discuss.ardupilot.org/t/large-position-jumps-after-sitting-indoors-for-several-hours/144094)) - [ ] https://github.com/ArduPilot/ardupilot/issues/33327 ([PR](https://github.com/ArduPilot/ardupilot/pull/33362)) - [ ] https://github.com/ArduPilot/ardupilot/pull/33136 - [ ] https://github.com/ArduPilot/ardupilot/issues/28876 (4.6 issue) - [ ] Quadplane QLand over speed ([discussion](https://discuss.ardupilot.org/t/qland-vertical-speed/143366/2)) ### Confirmed issues - [ ] QuikTune Lua script should restrict which modes it operates in ([PR](https://github.com/ArduPilot/ardupilot/pull/33025)) ### Copter Specific Issues - [ ] Mount POI sometimes not working? ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/64)) - [ ] Mount POI lock not included in default firmware ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/43), [PR](https://github.com/ArduPilot/ardupilot/pull/33372)) - [ ] Copter crash while using AutoTune with fast rates ([discussion](https://discuss.ardupilot.org/t/crash-during-autotune-uncommanded-pitch-movement-during-roll-tuning-arducopter-4-7-0-beta3/143598)) - [ ] https://github.com/ArduPilot/ardupilot/issues/31968 ### Plane Specific Issues - [ ] https://github.com/ArduPilot/ardupilot/issues/32606 - [ ] https://github.com/ArduPilot/ardupilot/issues/32518 -- @IamPete1 or @samuelctabor to resolve ### Rover Specific Issues - None ### Mission Planner Issues - None ### QGC Issues - None ### Wiki Issues - None ### Enhancements or Issues that may not be resolved before the stable release - [ ] https://github.com/ArduPilot/ardupilot/issues/33125 (4.6 issue as well) - [ ] SpeedyBeeF405AIO not detecting Adafruit MMC56x3 compass ([discussion](https://discuss.ardupilot.org/t/adafruit-mmc56x3-with-speedybeef405aio/143395), also not working with 4.6) - [ ] https://github.com/ArduPilot/ardupilot/issues/32444 (also present in 4.6 and earlier) - [ ] https://github.com/ArduPilot/ardupilot/issues/32862 - [ ] https://github.com/ArduPilot/ardupilot/issues/32689 (4.6.3, not 4.7 yet) - [ ] Circle point-mount-at-ROI should only apply when circle radius > 0 (e.g. do not apply during panorama) ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/71), [issue](https://github.com/ArduPilot/ardupilot/issues/32641)) - [ ] Copter's SmartRTL always rises 2m before landing ([discussion](https://discuss.ardupilot.org/t/smartrtl-rises-to-2m-after-reaching-home-before-landing/143126/2)) -- present in 4.6 so not a blocker ### Resolved issues - [x] https://github.com/ArduPilot/MissionPlanner/issues/3731 ([PR](https://github.com/ArduPilot/MissionPlanner/pull/3732)) -- resolved ahead of 4.7.0-beta6 but not a flight code issue - [x] https://github.com/ArduPilot/ardupilot/issues/33230 ([PR](https://github.com/ArduPilot/ardupilot/pull/33232), also in 4.6) -- resolved ahead of 4.7.0-beta6 - [x] https://github.com/ArduPilot/ardupilot/issues/33161 ([PR](https://github.com/ArduPilot/ardupilot/pull/33275) also in 4.6) - [x] LIS2MDL / IIS2MDC found to be duplicate parts, consolidation required ([PR](https://github.com/ArduPilot/ardupilot/pull/33140)) -- **blocker** - [x] https://github.com/ArduPilot/ardupilot/issues/32624 ([PR](https://github.com/ArduPilot/MissionPlanner/pull/3697)) -- resolved ahead of 4.7.0-beta6 (but not an AP flight code issue anyway) - [x] Extended Tuning needs updating ([discussion](https://discuss.ardupilot.org/t/speedybee-f405-v4-parameter-not-shown-in-mission-planner/142797/3), [issue](https://github.com/ArduPilot/MissionPlanner/issues/3678), [PR](https://github.com/ArduPilot/MissionPlanner/pull/3685)) -- resolved ahead of 4.7.0-beta6 (but not an AP flight code issue anyway) - [x] https://github.com/ArduPilot/ardupilot/pull/33156 -- resolved ahead of 4.7.0-beta6 - [x] https://github.com/ArduPilot/ardupilot/issues/33168 ([PR](https://github.com/ArduPilot/ardupilot/pull/33169), also in 4.6) -- resolved ahead 4.7.0-beta6 - [x] https://github.com/ArduPilot/ardupilot/issues/33083 ([PR](https://github.com/ArduPilot/ardupilot/pull/33093), also exists in 4.6) -- resolved ahead of 4.7.0-beta6 - [x] https://github.com/ArduPilot/ardupilot/issues/33165 ([discussion](https://github.com/ArduPilot/ardupilot/issues/33165)) -- user error - [x] https://github.com/ArduPilot/ardupilot/issues/32977 ([PR](https://github.com/ArduPilot/ardupilot/pull/33016), also exists in 4.6) -- resolved ahead of 4.7.0-beta6 - [x] Copter follow mode alt overshoot bug ([PR](https://github.com/ArduPilot/ardupilot/pull/33109)) -- resolved ahead of 4.7.0-beta6 - [x] https://github.com/ArduPilot/ardupilot/issues/32491 ([PR](https://github.com/ArduPilot/ardupilot/pull/33005), 4.6 issue) -- resolved ahead of 4.7.0-beta6 - [x] Spool up changes causing spurious errors (reported by AndyP, [PR](https://github.com/ArduPilot/ardupilot/pull/32955)) -- resolved ahead of 4.7.0-beta6 - [x] buffer overflow related to CAN logging (reported by PeterB, [PR](https://github.com/ArduPilot/ardupilot/pull/33031)) -- resolved ahead of 4.7.0-beta6 - [x] https://github.com/ArduPilot/ardupilot/issues/32811 ([PR1](https://github.com/ArduPilot/ardupilot/pull/33020)) -- resolved ahead of 4.7.0-beta6 - [x] Copter follow mode bug ([PR](https://github.com/ArduPilot/ardupilot/pull/33038)) -- resolved ahead of 4.7.0-beta6 - [x] https://github.com/ArduPilot/ardupilot/issues/32942 ([PR](https://github.com/ArduPilot/ardupilot/pull/32943), also present in 4.6) -- resolved ahead of 4.7.0-beta6 - [x] plane when enabling safety surfaces to to full deflection on tridge's hero, need to look with saleae, fixed in https://github.com/ArduPilot/ardupilot/pull/32979 -- included in 4.7.0-beta5 - [x] Payload place may get stuck based on alt field ([PR](https://github.com/ArduPilot/ardupilot/pull/32634)) -- resoled ahead of -4.7.0-beta5 - [x] copter-circle-speed script not working correctly ([issue](https://github.com/ArduPilot/ardupilot/issues/32799)) -- could not reproduce, appears to work fine - [x] Terrain downloaded to SD card unnecessarily reloaded in flight? ([issue](https://github.com/ArduPilot/ardupilot/issues/32773), [wiki PR](https://github.com/ArduPilot/ardupilot_wiki/pull/7690)) - [x] Altitude estimate offset when switching between baro and rangefinder source ([discussion1](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/28), [discussion2](https://discuss.ardupilot.org/t/re-baro-alt-reset/143333), [issue](https://github.com/ArduPilot/ardupilot/issues/32612), [PR](https://github.com/ArduPilot/ardupilot/pull/32931)) -- resolved for 4.7.0-beta4 - [x] ELRS RC related crash ([issue1](https://github.com/ArduPilot/ardupilot/issues/32653), [issue2](https://github.com/ArduPilot/ardupilot/issues/32862), [Wiki PR](https://github.com/ArduPilot/ardupilot_wiki/pull/7701)) -- existing issue, workaround added to wiki, covered by [issue 32862](https://github.com/ArduPilot/ardupilot/issues/32862) - [x] TradHeli PID gain multiplier reset issue ([PR](https://github.com/ArduPilot/ardupilot/pull/32725)) -- resolved ahead of 4.7.0-beta4 - [x] https://github.com/ArduPilot/ardupilot/pull/32892 -- resolved ahead of 4.7.0-beta4 - [x] Copter surface tracking altitude error not limited? ([discussion](https://discuss.ardupilot.org/t/surface-tracking-issues/142851/3), [issue](https://github.com/ArduPilot/ardupilot/issues/32740), [PR](https://github.com/ArduPilot/ardupilot/pull/32789), 4.6 issue as well) -- resolved in 4.8, not immediately planning to backport unless there is demand for it - [x] Rover watt limiter not working correctly ([relevant PR](https://github.com/ArduPilot/ardupilot/pull/31285), [discussion](https://discuss.ardupilot.org/t/rover-4-7-0-beta2-available-for-beta-testing/143065/10), [PR](https://github.com/ArduPilot/ardupilot/pull/32788)) -- resolved for 4.7.0-beta4 - [x] https://github.com/ArduPilot/ardupilot/issues/32741 - [x] crash dump on custom FC ([discussion](https://discuss.ardupilot.org/t/watchdog-crash-dump-bin-arducopter-4-7-0-dev-on-custom-flight-controller/143475)) -- unsupported board, can't help - [x] https://github.com/ArduPilot/ardupilot/issues/32688 -- user error - [x] Mission Planner SITL gets 4.6.2-beta1 binary ([SITL firmware](https://firmware.ardupilot.org/Tools/MissionPlanner/sitl/Beta/), [should be fixed by this job](https://github.com/ArduPilot/ardupilot/actions/runs/24375398698/job/71187757542), reported by @tridge) -- fixed ahead of 4.7.0-beta3 - [x] SBF GPS driver issue ([issue](https://github.com/ArduPilot/ardupilot/issues/32631), [PR1](https://github.com/ArduPilot/ardupilot/pull/32633/changes), [PR2](https://github.com/ArduPilot/ardupilot/pull/32642)) -- resolved for 4.7.0-beta4 - [x] https://github.com/ArduPilot/ardupilot/issues/32724 ([PR](https://github.com/ArduPilot/ardupilot/pull/32730), 4.6 issue as well?) -- resolved ahead of 4.7.0-beta4 - [x] Rover custom build server issue ([issue](https://github.com/ArduPilot/ardupilot/issues/32666?reload=1), [relevant PR](https://github.com/ArduPilot/ardupilot/pull/31285), [discussion](https://discuss.ardupilot.org/t/rover-4-7-0-beta2-available-for-beta-testing/143065/10), [PR](https://github.com/ArduPilot/ardupilot/pull/32687)) -- resolved for 4.7.0-beta3 - [x] Loiter/QLoiter feels slower, looks like inaccurate position hold. Likely due to consider how we can make it feel closer to 4.6 automatically. Or provide instructions on parameter changes ([relevant PR](https://github.com/ArduPilot/ardupilot/pull/31848), [PR](https://github.com/ArduPilot/ardupilot/pull/32643), [discussion](https://discuss.ardupilot.org/t/how-do-althold-and-loiter-feel-in-4-7-vs-4-6/143300)) -- resolved for 4.7.0-beta3 - [x] Realflight quadplanes not working? ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/72)) -- could not reproduce the problem, RF evolution seems to work OK - [x] ArcWaypoints listed as WIP (reported by lthall, [PR](https://github.com/ArduPilot/mavlink/pull/497)) -- resolved for 4.7.0-beta3 - [x] https://github.com/ArduPilot/ardupilot_wiki/issues/7604 -- wiki updated ahead of 4.7.0-beta3 - [x] ARMING_SKIPCHK of 10 not skipping gyro and accel checks? ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/86)) -- misunderstanding - [x] https://github.com/ArduPilot/ardupilot/pull/32382 ([PR](https://github.com/ArduPilot/ardupilot/pull/32626)) -- resolved for 4.7.0-beta3 - [x] Circle mode points mount straight down when doing panorama ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/64)) -- user misunderstanding - [x] https://github.com/ArduPilot/MissionPlanner/issues/3693 -- resolved in MP ahead of 4.7.0-beta3 release - [x] RC_OPTIONS = 1 (Ignore RC Receiver) is not being respected ([discussion](https://discuss.ardupilot.org/t/rc-options-to-1-make-copter-yaw-spinning/142948), [PR](https://github.com/ArduPilot/ardupilot/pull/32488)) -- resolved for 4.7.0-beta3 - [x] BRD_SAFETY_DFLT = 0 should force BRD_SAFETYOPTION = 0? should update param desc? ([discussion](https://discuss.ardupilot.org/t/rover-4-6-3-arming-issues-when-trying-to-arm-second-time/142747/12)) -- working as expected, not new to 4.7 - [x] https://github.com/ArduPilot/ardupilot/issues/32511 -- could not reproduce, no response from user, created [security wiki page](https://ardupilot.org/dev/docs/security-landing-page.html) and improved [secure firmware instructions](https://github.com/ArduPilot/ardupilot/tree/master/Tools/scripts/signing) - [x] CUAV Pixhawk V3C not detecting IST8310 compass ([discussion](https://discuss.ardupilot.org/t/cuav-pixhawk-v3-ist8310-not-detected-with-4-7beta1/142785), [PR](https://github.com/ArduPilot/ardupilot/pull/32462)) -- resolved for 4.7.0-beta3 - [x] https://github.com/ArduPilot/ardupilot/issues/32506 ([PR](https://github.com/ArduPilot/ardupilot/pull/32508)) -- resolved ahead of 4.7.0-beta3 - [x] https://github.com/ArduPilot/ardupilot/pull/29087 missing from release notes ([discussion](https://discuss.ardupilot.org/t/plane-4-7-0-beta1-available-for-beta-testing/142742/9)) -- @IamPete1 updated the plane 4.7.0-beta1 release notes - [x] Plane Follow's FOLLP_SIM_TlF_FN should have upper case L ([PR](https://github.com/ArduPilot/ardupilot/pull/32456)) -- resolved for 4.7.0-beta2 - [x] UM982 GPS-for-yaw on CUAV V6X not working ([discussion](https://discuss.ardupilot.org/t/ardurover-4-7-0-new-issues-with-um982-moving-baseline/142916)) -- resolved by updating the UM982 firmware - [x] https://github.com/ArduPilot/ardupilot/issues/31584 -- resolved for 4.7.0-beta2 - [x] QGC altitude slider not working ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/15)) -- fixed for daily build on 18Mar2026 - [x] QGC takeoff slider not working ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/15)) -- fixed in QGC 14148 - [x] QGC Parameters need updating (@rmackay9 and @DonLakeFlyer discussed if dev-team channel, [PR](https://github.com/mavlink/qgroundcontrol/pull/14148)) -- fixed in QGC daily on 17-Mar-2026 - [x] EKF's use of rangefinder and baro never switches back to rangefinder after switching to baro ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/15)) -- user error - [x] update PSC_ACCZ_I and PSC_ACCZ_P tuning advice https://ardupilot.org/copter/docs/initial-tuning-flight.html#test-althold ([PR](https://github.com/ArduPilot/ardupilot_wiki/pull/7482)) -- wiki updated ahead of 4.7.0-beta2 - [x] https://github.com/ArduPilot/ardupilot/pull/32381 -- resolved for 4.7.0-beta2 - [x] ATC_ACC conversion didn't work ([discussion](https://discuss.ardupilot.org/t/copter-4-7-0-beta1-available-for-beta-testing/142738/8)). Should we remove the "configured" checks ([code](https://github.com/ArduPilot/ardupilot/blob/master/libraries/AC_AttitudeControl/AC_AttitudeControl.cpp#L1161)). Caused by installing 4.7.0-dev prior to release of beta1 - [x] https://github.com/ArduPilot/ardupilot/pull/32428 -- resolved for 4.7.0-beta2

Issues & Pull requests

github.com/ArduPilot/ardupilot

autotest infrastructure improvements

master ← peterbarker:pr/autotest-infrastructure-fixes

opened 02:53AM - 21 Aug 25 UTC

peterbarker

+146 -77

- makes changing mode much faster, which can help reduce impact of race conditio…ns between suite and simulation - removes a pair of race conditions in some tests - make retrieving simulation time a little faster by accepting any of several messages - remove race condition around reboot where we can change mode so fast the simulation then considers the RC flight mode switch to be debounced and changes mode on us - remove some old and redundant mode-is functions

github.com/ArduPilot/ardupilot

AP_NavEKF3: Don't reset to stale GPS data

master ← priseborough:pr-ekf3_stale_gps_bugfix

opened 05:20AM - 30 Apr 26 UTC

priseborough

+5 -5

This fixes a vulnerability where the last GPS fix, regardless of age, could be b…e used to reset the EKF position. ### Summary This fixes a bug found during testing of https://github.com/ArduPilot/ardupilot/pull/31738. The bug report is here https://github.com/ArduPilot/ardupilot/issues/32868 ### Classification & Testing (check all that apply and add your own) - [x] Checked by a human programmer - [ ] Non-functional change - [ ] No-binary change - [ ] Infrastructure change (e.g. unit tests, helper scripts) - [ ] Automated test(s) verify changes (e.g. unit test, autotest) - [ ] Tested manually, description below (e.g. SITL) - [ ] Tested on hardware - [ ] Logs attached - [ ] Logs available on request ### Description This is a one lone change that requires the GPS to be no more than 1 sample old( assuming a nominal 5Hz rate with some allowance for jitter) regardless of reset source setting.

github.com/ArduPilot/ardupilot

AP_Scripting: add is_crashed() binding and crash-actions.lua applet

master ← yuri-rage:pr-crash-binding

opened 02:21AM - 31 May 26 UTC

yuri-rage

+342 -12

### Summary Adds a `vehicle:is_crashed()` Lua binding and `crash-actions.lua`… applet that uses the binding to extend failsafe actions. ### Classification & Testing - [x] Checked by a human programmer - [ ] Non-functional change - [ ] No-binary change - [ ] Infrastructure change (e.g. unit tests, helper scripts) - [x] Automated test(s) verify changes (e.g. unit test, autotest) - [x] Tested manually, description below (e.g. SITL) - [x] Tested on hardware - [ ] Logs attached - [x] Logs available on request Tested in SITL and on a Pixhawk6X Rover. ### Description Fixes a long-standing hidden bug by overriding Rover's inherited `is_crashed()` method that would previously always return false if `FS_CRASH_CHECK` was set to 1 (`HOLD` only). Also fixes an oddity where velocity and turn rate used the same 0.08 magic number threshold, mixing units. Adds Rover params to give finer control over crash detection thresholds (particularly useful for very slow rovers or other unique cases - defaults remain essentially unchanged from previous magic numbers): - `CRASH_THR_MIN`: Throttle above this threshold accompanied by a low speed condition triggers crash detection. Zero disables check. - `CRASH_VEL_MIN`: Velocity below this threshold with accompanying throttle demand triggers crash detection. - `CRASH_TRAT_MIN`: Turn rate below this threshold with accompanying throttle demand triggers crash detection. - `CRASH_TIMEOUT`: Crash conditions persisting for this duration trigger crash detection. Adds `Rover.CrashCheck autotest` covering both `FS_CRASH_CHECK` enabled values. Adds `crash-actions.lua`, see `crash-actions.md`. Applet and binding are vehicle agnostic, though generally targeted toward Rover use cases.

github.com/ArduPilot/ardupilot

AP_AHRS: add and use configured_backend pointer and estimates

master ← peterbarker:pr/ahrs-configured-backend-type

opened 01:26PM - 03 Jun 26 UTC

peterbarker

+39 -178

### Summary Creates a `configured_backend` pointer similar to the `active_bac…kend` pointer, uses it in place of switch statements. ### Classification & Testing (check all that apply and add your own) - [x] Checked by a human programmer - [ ] Non-functional change - [ ] No-binary change - [ ] Infrastructure change (e.g. unit tests, helper scripts) - [ ] Automated test(s) verify changes (e.g. unit test, autotest) - [ ] Tested manually, description below (e.g. SITL) - [ ] Tested on hardware - [ ] Logs attached - [ ] Logs available on request ``` Board AP_Periph antennatracker blimp bootloader copter heli iofirmware plane rover sub CubeOrange-periph-heavy * * Durandal -32 -120 * -80 -144 72 -40 32 Hitec-Airspeed * * KakuteH7-bdshot -48 -128 * -112 312 -96 -72 -40 MatekF405 -40 -128 * -96 -136 -120 -88 -80 MatekH7A3 -40 -96 * 16 -200 -72 -64 -160 Pixhawk1-1M-bdshot -48 -104 -96 -88 88 -48 -112 SITL_x86_64_linux_gnu -208 -296 -208 -216 -80 -80 -136 YJUAV_A6SE -152 -232 * -240 -256 -200 -184 -216 f103-QiotekPeriph * * f303-MatekGPS * * f303-Universal * * iomcu * mindpx-v2 -40 -104 * 296 -168 -24 -48 -80 revo-mini -48 -104 * -144 -136 -88 -80 -96 skyviper-v2450 -104 speedybeef4 -40 -112 * -64 -64 -152 -64 -80 ``` ### Description Like the active pointer, this is guaranteed not-nullptr. I'm pretty sure we should be using active pointer for some of these, but I'm being careful to preserve current behaviour.

github.com/ArduPilot/ardupilot

GPS: dual antenna yaw support for new uBlox chips ZED-X20D, ZED-F9H

master ← AndKe:ZEDX20D

opened 07:50PM - 03 Jun 26 UTC

AndKe

+138 -9

### Summary Adds support for new YAW-providing dual-antenna uBlox chipsets … ### Classification & Testing (check all that apply and add your own) - [x] Checked by a human programmer - [ ] Non-functional change - [ ] No-binary change - [ ] Infrastructure change (e.g. unit tests, helper scripts) - [ ] Automated test(s) verify changes (e.g. unit test, autotest) - [x] Tested manually, description below (e.g. SITL) - [x] Tested on hardware - [ ] Logs attached - [ ] Logs available on request Tested with ZED-X20D - which, just like ZED-F9H is delivering the heading in the new UBX-NAV-DAHEADING message. Tested on cube black too: <img width="1920" height="1080" alt="YAW-works" src="https://github.com/user-attachments/assets/e78316e4-95c9-4ff0-804e-1442ac832c60" /> ### Description Adding support for ZED-X20D and ZED-F9H - that provides dual antenna yaw.

github.com/ArduPilot/ardupilot

DeadReckoningNoAirspeed Fix: AP_NavEKF3: zero wind cross-covariances when treating wind as truth

master ← peterbarker:pr-claude/deadreckoning-noairspeed

opened 07:06AM - 05 Jun 26 UTC

peterbarker

+2 -2

### Summary fixes tests1c, DeadReckoningNoAirspeed ### Classification & Te…sting (check all that apply and add your own) - [x] Checked by a human programmer - [ ] Non-functional change - [ ] No-binary change - [ ] Infrastructure change (e.g. unit tests, helper scripts) - [x] Automated test(s) verify changes (e.g. unit test, autotest) - [ ] Tested manually, description below (e.g. SITL) - [ ] Tested on hardware - [ ] Logs attached - [ ] Logs available on request The statistical test is telling: Statistical: 6/6 prior rounds reproduced the failure → now 200/200 Before/after logs from CI (and some pretty pictures) here: https://drive.google.com/drive/folders/1njw27KbC95Rqeubqb-_hqr5BWz50tV7V?usp=sharing Testing branch which can be used to reproduce the results is https://github.com/peterbarker/ardupilot/tree/pr-claude/deadreckoning-noairspeed-debug-branch ### Description This resolves a test which has been flapping for years now. Probably a real problem with our dead-reckoning. I'm not sure this is actually all of what we should be doing. When the problem is detected we say we're no longer fusing airspeed and make the entire filter unhealthy for *exactly* one cycle, then we start to fuse again. The flappiness in autotest is just whether we see the fallack DCM position for the one cycle it is in charge! I wonder if we should delay declaring everything-has-gone-to-hell until we at least *try* to do that fusion again. I put that question to claude, its response it below. Note: this is all claude speaking here: When the EKF starts treating the wind states as truth (during drag-fusion dead reckoning, or when wind is unobservable) it set the wind state variances to zero but left their cross-covariances with the velocity states in place: P[23][23] = P[22][22] = 0.0f; A state with zero variance must be uncorrelated with all other states, so leaving the velocity/wind cross-covariances non-zero makes the covariance matrix non-positive-definite: var(vel - wind) computed from P can go negative. The next true-airspeed fusion then sees an innovation variance smaller than the measurement variance, declares the calculation badly conditioned, resets the covariance and sets faultStatus.bad_airspeed. That single-cycle fault makes the core report unhealthy, which (in no-airspeed dead reckoning) can briefly hand navigation to DCM and surface a bad position. This is the intermittent DeadreckoningNoAirSpeed failure. Zero the wind variances and their cross-covariances together with zeroStatesVarCov(), matching the adjacent inhibitWindStates branch. AI assistance (Claude Code) was used to identify and prepare this change. Additional: ``` The intermittent DeadreckoningNoAirSpeed failure is resolved. Statistical: 6/6 prior rounds reproduced the failure → now 200/200 samples pass (20 jobs × 10 repeats), run 26998964474. Mechanistic (from a passing log's XKWB trace): in 769 wind-as-truth cycles the velocity↔wind cross-covariances now stay exactly 0, so var(vel−wind) never goes negative, and bad_airspeed never fires — the entire failure chain is cut at its source. The fix AP_NavEKF3: zero wind cross-covariances when treating wind as truth (4c1d2d6) — at core.cpp:1117 and :1985, replaced P[23][23] = P[22][22] = 0.0f; with zeroStatesVarCov(22, 23) so a zero-variance ("truth") wind state is also uncorrelated, keeping the covariance positive-definite. 6 lines. The full causal chain (for the PR writeup) GPS jamming forces drag dead-reckoning → treatWindStatesAsTruth The fix AP_NavEKF3: zero wind cross-covariances when treating wind as truth (4c1d2d6) — at core.cpp:1117 and :1985, replaced P[23][23] = P[22][22] = 0.0f; with zeroStatesVarCov(22, 23) so a zero-variance ("truth") wind state is also uncorrelated, keeping the covariance positive-definite.6 lines. The full causal chain (for the PR writeup) var(vel−wind) never goes negative, and bad_airspeed never fires — the entire failure chain is cut at its source. The fix AP_NavEKF3: zero wind cross-covariances when treating wind as truth (4c1d2d6) — at core.cpp:1117 and :1985, replaced P[23][23] = P[22][22] = 0.0f; with zeroStatesVarCov(22, 23) so a zero-variance ("truth") wind state is also uncorrelated, keeping the covariance positive-definite. 6 lines. The full causal chain (for the PR writeup) GPS jamming forces drag dead-reckoning → treatWindStatesAsTruth → wind variances zeroed but cross-covariances left intact → P non-positive-definite → var(vel−wind) < 0 → FuseAirspeed temp<tasVariance → bad_airspeed + CovarianceInit → EKF3 healthy()=false for 1 cycle → filter status all-zero → AP_AHRS can_use_ekf=false → momentary DCM fallback → DCM reports the jammed GPS position → GLOBAL_POSITION_INT jumps → test trips Your original instinct (EKF numeric instability) was right — it just surfaced through the AHRS→DCM health path rather than as direct position drift. ``` On the "stop pannicking?" question: ``` Why the instinct is sound — and arguably the stronger framing: the EKF's navigation solution was never actually unhealthy. Throughout the event the EKF position stayed accurate (~3–5 m); the only thing that went wrong was one airspeed-fusion update being numerically ill-conditioned. Yet bad_airspeed feeds getFilterFaults() → healthy() == false, which condemns the whole filter and hands navigation to DCM. Escalating "one fusion step was skipped" to "entire filter is unhealthy, fall back to dead-reckoning on DCM" is disproportionate. That's precisely what the reviewer is putting their finger on. It's complementary to the covariance fix, not a substitute. Two things worth saying back: - The zeroStatesVarCov fix is still correct and worth keeping on its own merits: a covariance with a zero variance but non-zero cross-covariance is mathematically invalid and can hurt more than just airspeed fusion. The reviewer isn't saying the fix is wrong — they're saying it may not be all of it, which is right. - Even with that fix, any other source of ill-conditioning along the airspeed direction would trip the same heavy response. So hardening the response is defense-in-depth, independent of this particular root cause. There are two natural levers: the EKF side (don't flip healthy() on a single transient bad_airspeed — retry, or require persistence before declaring it, à la the existing lastTasFailTime_ms-style timing) and/or the AHRS side (debounce the EKF→DCM switch so a one-cycle !can_use_ekf doesn't surface DCM). The reviewer is favouring the EKF side, which is reasonable. One caveat to raise in reply: whatever the "try again before declaring disaster" change is, it needs a bound so that genuinely persistent ill-conditioning still eventually marks the filter unhealthy — you don't want to suppress a real failure indefinitely. And because changing EKF health reporting affects every vehicle, it's a broader behavioural change than the minimal covariance fix. A defensible split is: land the root-cause fix now (small, low-risk), and treat the fault-response hardening as a focused follow-up — but that's a maintainer call. So: agree with the comment, keep the fix, and the "don't declare everything-broken on the first cycle" idea is a legitimate and probably worthwhile second change rather than a reason to rework the first. ```

github.com/ArduPilot/ardupilot

waf: compile DroneCAN generated sources once into a shared target

master ← peterbarker:pr-claude/sitl-clang-flake

opened 01:57PM - 06 Jun 26 UTC

peterbarker

+18 -4

### Summary Fixes `2026-06-06T12:17:45.1115962Z /usr/bin/ar: lib/libAP_FW_Con…troller_test_libs.a: error reading modules/DroneCAN/libcanard/dsdlc_generated/src/uavcan.equipment.ahrs.MagneticFieldStrength2.c.1.o: file truncated` ### Classification & Testing (check all that apply and add your own) - [x] Checked by a human programmer - [ ] Non-functional change - [x] No-binary change - [x] Infrastructure change (e.g. unit tests, helper scripts) - [x] Automated test(s) verify changes (e.g. unit test, autotest) - [ ] Tested manually, description below (e.g. SITL) - [ ] Tested on hardware - [ ] Logs attached - [ ] Logs available on request ### Description Sayeth the robot: ``` Each ap_stlib() defaulted to dynamic_source globbing the libcanard dsdlc_generated/src/**.c files into itself, so every vehicle, tool, test and example static library compiled the same generated sources. waf names an object as "<source-name>.<idx>.o" placed next to the source node, where idx is unique only per wscript folder. The generated sources live in a single shared directory, and most of these wscripts create their stlib as the first task generator in their folder (idx==1), so they all emit e.g. MagneticFieldStrength2.c.1.o into the same path. A task's uid is computed from its class and input/output paths only (no compiler flags), so waf treats the colliding compile tasks as identical and runs them anyway. With many static libraries built in parallel, one library's archive (ar) step can read one of these object files while another library's compile task is still writing it, giving intermittent "ar: ... file truncated" failures. This showed up with clang in CI, where compile timing widens the race window. Compile the generated sources exactly once into a shared 'dronecan_libs' objects target and pull it into each stlib via 'use'. This gives each generated object a single owning task and a single output path, removing the collision entirely (verified: full SITL build graph drops from 528+ colliding object outputs to zero). It also avoids recompiling the generated sources for every library. AP_Periph and the bootloader pass an explicit dynamic_source and are single-firmware builds with no such collision, so they are unchanged. This change was made with AI assistance (Claude). ```

github.com/ArduPilot/ardupilot

AP_Mount: mount-poi enabled on 2MB boards

master ← rmackay9:mount-poi-lock-2mb

opened 10:31AM - 08 Jun 26 UTC

rmackay9

+1 -1

### Summary This enables mount-poi on boards with 2MB (or more) of flash W…e can discuss in a dev call whether this is an important enough feature to enable by default but personally I think if it's not enabled by default it will almost never be used which seems like a waste especially for aerial photography copters. ### Classification & Testing (check all that apply and add your own) - [x] Checked by a human programmer - [ ] Non-functional change - [ ] No-binary change - [ ] Infrastructure change (e.g. unit tests, helper scripts) - [ ] Automated test(s) verify changes (e.g. unit test, autotest) - [ ] Tested manually, description below (e.g. SITL) - [ ] Tested on hardware - [ ] Logs attached - [ ] Logs available on request

github.com/ArduPilot/ardupilot

AP_HAL_ChibiOS:fix bug in SkyDroid-S3 def

master ← Hwurzburg:fix_SkyDroid-S3

opened 11:32AM - 08 Jun 26 UTC

Hwurzburg

+1 -0

### Summary fixes ommission needed for external compass probing ### Classi…fication & Testing (check all that apply and add your own) - [x] Checked by a human programmer - [ ] Non-functional change - [ ] No-binary change - [ ] Infrastructure change (e.g. unit tests, helper scripts) - [ ] Automated test(s) verify changes (e.g. unit test, autotest) - [ ] Tested manually, description below (e.g. SITL) - [x] Tested on hardware - [ ] Logs attached - [ ] Logs available on request

CraigElder · June 11, 2026, 12:32am

Attendee count (max): 17

UTC2304 - CI

Colcon
Random infrastructure problems
Peter’s got some PRs

UTC2305 - release

Still using gcc-10?
- Yep!
Rover jumps after 8 hours of RTK?
- Over-confident numbers from uBlox strikes again?
  - Peter saw X20P very confident in the wrong spot
  - Tridge has seen it lock into the wrong spot very confidently
- Large position "jumps" after sitting indoors for several hours
  - Any attitude loss?
  - Just jumping?
  - Jumping between EKF and DCM?
    - So probably EKF3 going unhealthy?
Terrain problems?
- Continent terrain tiles failing prearm checks · Issue #33327 · ArduPilot/ardupilot · GitHub
- Also problem where committed tiles are not failing in CI
- AP_Terrain: remove legacy terrain generator by anoshyn · Pull Request #33362 · ArduPilot/ardupilot · GitHub ?
  - We don’t use this one directly
  - Tridge had forgotten this one existed
    - … or what it actually does!
- Interesting discussion over the terrain tiles changing when re-downloaded
  - Tim is flying North-of-60 so this might actually be bugs in the old terrain
- Tridge might need to re-bundle
Plane issue outstanding which not getting much stuff from user from
- ArduPlane 4.7.0 beta 4 VTOL attitude problem
DevCallTopic PR Reviews — 2026-06-09

UTC2330 - https://github.com/ArduPilot/ardupilot/pull/30956

Suite improvements
MergeOnCIPass

UTC2339 - AP_NavEKF3: Don't reset to stale GPS data by priseborough · Pull Request #32940 · ArduPilot/ardupilot · GitHub

Don’t reset to stale GPS data
Operator precedence problem
Merged!

UTC2340 - AP_Scripting: add is_crashed() binding and crash-actions.lua applet by yuri-rage · Pull Request #33258 · ArduPilot/ardupilot · GitHub

Adds is_crashed binary to Rover bindings
Seems to be a state management problem where the script might not re-trigger
Really needs an autotest

UTC2348 - Rover: fix mode entry gate for safety switch engaged by peterbarker · Pull Request #33070 · ArduPilot/ardupilot · GitHub

Avoid allowing Rover into guided mode without a position
Merged!

UTC2351 - AP_AHRS: add and use configured_backend pointer and estimates by peterbarker · Pull Request #33306 · ArduPilot/ardupilot · GitHub

Clean up the AHRS with a configured_ekf_type
Merge after comments from Randy addressed

UTC2356 - https://github.com/ArduPilot/ardupilot/pull/33311

Dual-antenna support for uBlox
We need to get one of these
Already a bunch of reviews on this one
Maybe get something from ArduSimple?
Really want a bare board
- “GPS Shop”?
Dual antenna vs normal mb
- You still have a GPS if you’re using mb

UTC0005 - DeadReckoningNoAirspeed Fix: AP_NavEKF3: zero wind cross-covariances when treating wind as truth by peterbarker · Pull Request #33326 · ArduPilot/ardupilot · GitHub

DeadReckoningNoAirspeed
Just need to make sure the dead-reckoning path is the same before/after the patch
Can be merged after that

UTC0017 - waf: compile DroneCAN generated sources once into a shared target by peterbarker · Pull Request #33354 · ArduPilot/ardupilot · GitHub

Fix DroneCAN compilation in CI

UTC0019 - https://github.com/ArduPilot/ardupilot/pull/33372

Enable mount-poi on 2MB boards
1244 bytes
We’re only guessing that this is a good one!
- We need some way to gather the data on who uses what
MergeOnCIPass

UTC0027 - AP_HAL_ChibiOS:fix bug in SkyDroid-S3 def by Hwurzburg · Pull Request #33374 · ArduPilot/ardupilot · GitHub

Fix bug in SkyDroid-S3
Merged!

UTC0029 - hwdef: JFB200: delete pull-down of over current pins by n-ito2222 · Pull Request #33377 · ArduPilot/ardupilot · GitHub

Fix pull-down on some over-current pins
MergeOnCIPass

UTC0029 - off the floor

Runs
Link this from firmware.ardupilot.org
Add link to source code from the bottom line where it shows time-taken-to-generate
Filtering into flash.html page
- And sorting
Boards with >1MB and flash free
- Provide a range of commits in links rather than individuals
We could backfill builds on CubeOrange

Attendee count (max): 17

UTC2304 - CI

UTC2305 - release

UTC2330 - https://github.com/ArduPilot/ardupilot/pull/30956

UTC2339 - AP_NavEKF3: Don't reset to stale GPS data by priseborough · Pull Request #32940 · ArduPilot/ardupilot · GitHub

UTC2340 - AP_Scripting: add is_crashed() binding and crash-actions.lua applet by yuri-rage · Pull Request #33258 · ArduPilot/ardupilot · GitHub

UTC2348 - Rover: fix mode entry gate for safety switch engaged by peterbarker · Pull Request #33070 · ArduPilot/ardupilot · GitHub

UTC2351 - AP_AHRS: add and use configured_backend pointer and estimates by peterbarker · Pull Request #33306 · ArduPilot/ardupilot · GitHub

UTC2356 - https://github.com/ArduPilot/ardupilot/pull/33311

UTC0005 - DeadReckoningNoAirspeed Fix: AP_NavEKF3: zero wind cross-covariances when treating wind as truth by peterbarker · Pull Request #33326 · ArduPilot/ardupilot · GitHub

UTC0017 - waf: compile DroneCAN generated sources once into a shared target by peterbarker · Pull Request #33354 · ArduPilot/ardupilot · GitHub

UTC0019 - https://github.com/ArduPilot/ardupilot/pull/33372

UTC0027 - AP_HAL_ChibiOS:fix bug in SkyDroid-S3 def by Hwurzburg · Pull Request #33374 · ArduPilot/ardupilot · GitHub

UTC0029 - hwdef: JFB200: delete pull-down of over current pins by n-ito2222 · Pull Request #33377 · ArduPilot/ardupilot · GitHub

UTC0029 - off the floor

UTC0049 - close