Using the SF30/d with the legacy protocol caused a delay of the
measurements of ~1s. This is not the case with the binary protocol anymore.
The initialization sequence used for SF20/c did not work and is therefore
updated.
Tested on both SF20/c and SF30/d.
This allows external modes to individually check if they are flagged as
invalid/unresponsive.
Previously this was done only based on whether or not ArmingCheckRequest
was received, which does not work when multiple modes are running.
based on the message timestamp.
Previously it was possible to run into the following case:
- 2 external modes are registered (running inside the same ROS node)
- they time out due to the micro xrce agent being blocked for some reason
- PX4 removes them
- the latest arming check replies still arrive to PX4
- the application restarts
- the first mode gets registered
- PX4 handles the previous arming check reply, and clears
waiting_for_first_response, which reduces the timeout
- the second mode registers and as part of that checks for message
compatibility. This takes ~1s, triggering a timeout of the first mode
There were a number of cases where the state was not correct or not as
desired after disarming, when running an external mode 'MyMission' with
executor:
- run MyMission, which triggers Hold, then Land
- before: Mode: Hold, executor_in_charge: 1
- after: Mode: MyMission, executor_in_charge: 1
- run MyMission, then user switches to RTL
- before: Mode: MyMission, executor_in_charge: 0
- after: Mode: MyMission, executor_in_charge: 1
- run MyMission, then while in Hold mode, low battery failsafe (RTL)
- before: Mode: Hold, executor_in_charge: 1
- after: Mode: MyMission, executor_in_charge: 1
- run MyMission, then stop external mode (terminate the process)
- before: Mode: (mode not available), executor_in_charge: 0
- after: Mode: Hold, executor_in_charge: 0
This case is unchanged:
- run MyMission, then low battery failsafe (RTL)
- before: Mode: MyMission, executor_in_charge: 1
- after: Mode: MyMission, executor_in_charge: 1
There was a build error with the previous container:
CMake Error at /usr/share/cmake-3.28/Modules/FindPackageHandleStandardArgs.cmake:230 (message):
Could NOT find Threads (missing: Threads_FOUND)
Call Stack (most recent call first):
/usr/share/cmake-3.28/Modules/FindPackageHandleStandardArgs.cmake:600 (_FPHSA_FAILURE_MESSAGE)
/usr/share/cmake-3.28/Modules/FindThreads.cmake:226 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
build/px4_sitl_test/_deps/abseil-cpp-src/CMakeLists.txt:98 (find_package)
See https://github.com/PX4/PX4-Autopilot/actions/runs/16235116238/job/45844179833
Previously, when switching from a goto setpoint into a mode that publishes
trajectory_setpoint, the previous goto setpoint was still used for 500ms,
which then caused a setpoint jump.
This change makes sure that when a trajectory_setpoint is received, any
existing goto setpoint is marked as invalid immediately.
There was a race condition: for example when an external mode disabled
failsafe deferring, that then triggered a failsafe, while the mode executor
immediately sends a command (to e.g. switch modes).
In that case the failsafe got triggered but the mode switch was still
allowed.
This was because of the processing ordering:
- mode updates (and propagating the failsafe_action_active state)
- failsafe updates
- command handling
This patch makes sure failsafe_action_active is set immediately after
updating the failsafes.
Previously, when deferring was active and e.g. RC loss was triggered, and
RC regained, the action was not cleared, as the RC loss action only clears
on mode switch/disarm (when set to RTL for example).
When deferring was then disabled, the RC loss failsafe would still trigger.
This changes the behavior to immediately remove those actions when
deferring is active.
It also ensures to reset the Hold delay when deferring is disabled and no
failsafe is being deferred.
Even though all enum values fit into 13 bits, there's an error:
../../src/drivers/gnss/septentrio/sbf/messages.h:101:21: error: ‘septentrio::sbf::Header::id_number’ is too small to hold all values of ‘enum class septentrio::sbf::BlockID’ [-Werror]
This is with GCC 13.3.0 under Ubuntu 24.04
From https://github.com/google/fuzztest.
This will now also add gtest (via cmake FetchContent)
And requires newer cmake (container updates):
CMake 3.19 or higher is required. You are running version 3.16.3
There was a race condition when closing the shell:
- the main thread checks if _mavlink_shell is not nullptr (which is true)
- the receiver thread closes the shell, which clears _mavlink_shell
- the main thread continues with _mavlink_shell->available()
_message.header.length - 4 is passed as unsigned to the CRC method, so if
_message.header.length < 4, the length wraps and causes invalid memory
access.
There is some race condition where in rare cases the topic publication
right after creating the writer did not get received on the ROS side.
This happens even with reliable QoS & reliable transport.
There is already another check for battery_unhealthy, so a separate state
and ID are required.
Fixes the error:
ERROR [failsafe] BUG: duplicate check for caller_id 74
We had a setup where the voltage was right at the threshold and the check
toggled continuously.
It still triggers immediately, and then keeps for 15 seconds
As the failsafe message can reference the health report, the health report
needs to be sent out first. This is generally the case, except there is a
rate limiter set to 2 seconds. So if the report changes quickly, it is
sent out delayed (potentially after the failsafe report).
We've come accross a case where a ROS node would consistently take something
over 800 ms until the first arming check request subscription callback was
triggered.
After the first sample, the callback always triggered within the expected
timeframe.
Therefore this patch allows for more time right after registration until
timing out.
192.168.0.x is often used by routers for WIFI/ethernet networks, and thus
can create conflicts.
This can happen for example if a companion is connected to the FMU via
ethernet and at the same time connects to a WIFI network as DHCP client.
Previously, if a mag was not required (not index 0 and not used by ekf),
it was not counted in num_enabled_and_valid_calibration.
If a user set SYS_HAS_MAG to e.g. 3, it would then trigger a preflight
failure, even if there were 3 calibrated and enabled mags.
Regression from https://github.com/PX4/PX4-Autopilot/pull/23043
Also avoids a race condition by making sure the command ack is handled
before sending out the mavlink message (in case an external component
reacts immediately to the mavlink message).
When handling multiple commands, it could happen that the first command
updates _reposition_triplet. Normally this would then get handled after
getting the mode change from commander through vehicle_status.
But if the next command is handled before an update from commander, it
could overwrite the triplet.
This patch ensures that navigator waits for an update from commander (and
therefore process the _reposition_triplet) before handling the next
command.
This happened specifically when pressing 'Pause' from QGC during a mission.
QGC sends VEHICLE_CMD_DO_REPOSITION twice, first for pausing, then changing
the altitude.
The result was that the vehicle would not stop at the current location but
continue to the next mission waypoint and stop there.
It was failing occationally with:
14:12:43 CMake Error at CMakeLists.txt:129 (list):
14:12:43 list index: 1 out of range (-1, 0)
14:12:43
14:12:43
14:12:43 CMake Error at CMakeLists.txt:131 (list):
14:12:43 list index: 2 out of range (-1, 0)
14:12:43
14:12:43
14:12:43 CMake Error at CMakeLists.txt:132 (string):
14:12:43 string sub-command REPLACE requires at least four arguments.
14:12:43
14:12:43
14:12:43 -- PX4 version: f513c40d62 (f513c40d62..NOTFOUND)
Reduces flash usage by ~16KB.
- compress formats at build-time into a single string with all formats
- then at runtime iteratively decompress using
https://github.com/atomicobject/heatshrink
The length check for unknown packets did not include PACKET_SIZE_TYPE_SIZE
and CRC_SIZE, and hence working_index could overflow CRSF_MAX_PACKET_LEN,
triggering invalid memory access further down in QueueBuffer_PeekBuffer.
Also the working_segment_size was wrong for unknown packets.
Credits for finding this go to @Pwn9uin.
