Repository navigation
Add Tracker tracking test - #17783
peterbarker wants to merge 6 commits into
Conversation
71c420a to
1acc044
Compare
9bbac19 to
614bf73
Compare
The limits tested yaw_rate and pitch_rate before update_onoff_servos() or update_position_servos() had assigned them, so they only ever saw the zeroes the two were initialised to and never fired. The simulated tracker could slew past its own yaw_range and pitch_range without limit. Move the block below the servo update, which is what sets the rates.
A PWM of zero means the channel is not being output to at all, and that is what the tracker emits whenever its servos are not being driven. Feeding it through (pwm-1500)/500 turns it into a demand of -3.0, which with yaw_range 170 and pitch_range 45 asks for -510 degrees of yaw and -135 of pitch, so the model slews at its rate limit for as long as nothing drives the servos. A real servo with no signal holds position, and the rest of SITL already treats a zero PWM that way - see Aircraft::filtered_servo_angle(). Do the same here: take no new demand from an undriven channel and leave its rate at zero. The slew this removes was not harmless. The tracker's simulated yaw swept at its 9 deg/sec limit from the first physics step, and because a yaw about the gravity vector does not disturb the accelerometers, the startup gyro calibration's only movement check could not see it: it converged and stored INS_GYROFFS_Z of -0.157 rad/sec, exactly the slew rate. The EKF was then fed a gyro denying a rotation its compass could see, drove the Z gyro bias state out to 8.8 deg/sec to reconcile them, and so never satisfied delAngBiasLearned - leaving readyToUseGPS() false and the filter unable to begin absolute aiding.
614bf73 to
c7d4295
Compare
Global Size Summary (Total Flash delta in bytes)Built from dd54cb3 rebased onto dc78bc5
|
Previous review (2026-10-18)Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting. Reviewed at head Thanks, this looks good, and nothing blocks. I rebuilt SITL at c7d4295 and Tracking passes. Earlier runs also showed it passing in suite order and at speedup 5. It fails when the firmware's bearing or baro altitude is deliberately offset, and when the position hook is removed, so it does catch real pointing errors. One small comment fix: Tools/autotest/antennatracker.py:253 (https://github.com/ArduPilot/ardupilot/pull/17783/files#diff-d7d46e38a9dc3898d10c1828f0c345cab42344ad52194eca21eb2f12fc1beeefR253) says the simulated tracker "starts pointing west", and the docstring at line 146 (https://github.com/ArduPilot/ardupilot/pull/17783/files#diff-d7d46e38a9dc3898d10c1828f0c345cab42344ad52194eca21eb2f12fc1beeefR146) says it starts at 270 degrees yaw and 10 degrees pitch. It actually boots at the SITL home heading (about 9 degrees) with level pitch; 270/10 is the centre of its +/-170 yaw and +/-45 pitch range. The test geometry is still correct. Something like "the simulated tracker's yaw range is centred on west and spans 170 degrees either way" would be accurate. The older SIM_Tracker.h comment has the same slip. For information: this carries #34733's three commits unchanged and Tracking needs them. It fails arming checks with the old simulator model. A note in the description that it builds on #34733 might help whoever merges. |
c7d4295 to
55a6f23
Compare
peterbarker
moved this from ReadyForDevCall
to In Progress
in Peter's ArduPilot 4.8 Queue
UndrivenServosHoldPosition centres the tracker, disarms so that both servo outputs read zero, and checks the simulated attitude does not move. Without the undriven-channel fix in SIM_Tracker the model yaws more than 100 degrees in the 12 seconds checked. ServoRangeLimits drives both servos to 2200 with DO_SET_SERVO, asking for 1.4 times the model's yaw and pitch range, and checks it stops and holds at +170 yaw and +45 pitch. Without the limits being applied after the servo update the model never settles there.
zero_yaw and zero_pitch are the centre of the tracker's yaw and pitch ranges, which a centred position servo points it at. The tracker does not start there: it boots at the SITL home heading with level pitch.
desyaw is compared using heading_delta so that it wraps correctly. minimum_duration requires the attitude to be held for that many seconds rather than returning on the first message within tolerance. It defaults to None, so existing callers keep their current behaviour, and must be less than timeout or the wait could never succeed.
Feed AUTO mode a vehicle position at a series of bearings from the tracker and check it points at each one and holds there. The vehicle reports the tracker's own pressure so the tracker should hold zero pitch throughout.
55a6f23 to
dd54cb3
Compare
|
Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting. Reviewed at head Thanks for the fixes. The head moved from c7d4295 to dd54cb3, and the change is comment-only. The docstring at Tools/autotest/antennatracker.py:146 (https://github.com/ArduPilot/ardupilot/pull/17783/files#diff-d7d46e38a9dc3898d10c1828f0c345cab42344ad52194eca21eb2f12fc1beeefR146), the Tracking comment at Tools/autotest/antennatracker.py:254 (https://github.com/ArduPilot/ardupilot/pull/17783/files#diff-d7d46e38a9dc3898d10c1828f0c345cab42344ad52194eca21eb2f12fc1beeefR254) and libraries/SITL/SIM_Tracker.h:46 (https://github.com/ArduPilot/ardupilot/pull/17783/files#diff-4572d7ac789c0553dda5d05f738a191864606beb25f1218981567f6066976a1dR46) now all correctly describe 270/10 as the centre of the range. I rebuilt SITL at dd54cb3: the two servo tests and Tracking pass, and Tracking also passes at speedup 5. The full Tracker suite passed in an earlier pass at this head. Nothing blocks. Two small things in the description, for information only. #34733 now has four commits, not three, and this PR carries all four. test.Tracker has 23 runnable tests rather than 24; all of them pass. The extra 'PASSED' in the log is GPSForYaw's result echoed in a later message. The first one goes away when you rebase after #34733 merges. CI for this head was still running when checked, with no failures so far. |
peterbarker
moved this from In Progress
to ReadyForDevCall
in Peter's ArduPilot 4.8 Queue
4 participants
Summary
Adds an autotest checking that Tracker in AUTO mode points at a vehicle sending GLOBAL_POSITION_INT and holds there.
Classification & Testing (check all that apply and add your own)
SITL:
test.Tracker: all 24 tests pass, including the newTrackingtest.Copter.ModeFlipandtest.Plane.ThrottleFailsafepass with thewait_attitudechange; they are among its existing callersflake8andTools/scripts/check_branch_conventions.pyare cleanIn
Tracking, each vehicle position starts at least 45 degrees from where the tracker is pointing, the first 100 degrees away, so none of the cases is satisfied by the tracker's starting attitude.Description
autotest: allow wait_attitude to wait for yaw and for a held attitudeaddsdesyaw, compared usingheading_deltaso it wraps, and an optionalminimum_durationthe attitude must be held for. Both default to off, so existing callers are unchanged. Aminimum_durationthat is not less thantimeoutis rejected, as the wait could never succeed.autotest: add test for tracker pointing at a vehicleaddsTracking. It feeds the tracker a vehicle position at 10Hz from a message hook (Tracker drops a target after 5 seconds without an update), places the vehicle at five bearings to the west, and requires the tracker to hold within 2 degrees for 5 seconds at each. The vehicle reports the tracker's own pressure, so the tracker should hold zero pitch throughout.This is built on #34733: the first three commits are that PR. Without its fixes the simulated tracker slews while its servos are undriven, the startup gyro calibration absorbs the slew, and the EKF never becomes ready for the test to start. I'll rebase once #34733 is merged.
This PR was first opened in 2021 and has been reworked. The PID logging it used to carry is now in #17778, and its changes to the simulated tracker's yaw range and its default parameter file have been dropped.
This contribution was AI-assisted: the rework of the original branch and the new test code were written with Claude Code and reviewed by me.