- matrix in PX4/Firmware (c475bb84b7274d041065fc3fd673127ae2438f4d): 9f46483951
- matrix current upstream: 93d42947b6
- Changes: 9f46483951...93d42947b693d4294 2019-11-08 kritz - Add test for setting Quaternion to Identity (#104)
- GPSDrivers in PX4/Firmware (085bdd14b41ac3977d612a1cae27f111de7fe4fb): 011959b4da
- GPSDrivers current upstream: 085a85c48a
- Changes: 011959b4da...085a85c48a
085a85c 2019-10-15 Andreas Antener - sbf: invalidating gps position when invalid data is received We have encountered a case where do-not-use values were being reported in velocity fields without a corresponding error code or fix-type 0. This adds a check for known invalid data and sets the appropriate flags.
Main UAVCAN protocol handling and ESC updates run on the same thread/wq as
before. There are 2 WorkItems for separate scheduling of the 2, so that
ESC updates run in sync with actuator_control updates. UAVCAN is scheduled
at a fixed rate of 3ms (previously the poll timeout) and on each UAVCAN
bus event.
This leads to roughly the same behavior as before. CPU & RAM usage are
pretty much the same (tested on Pixhawk 4).
Testing done: Motors still work (with feedback), param changes and a
UAVCAN optical flow sensor.
esc_setpoint in UAVCAN was just wrong, this is what it really is:
uint7 power_rating_pct # Instant demand factor in percent
(percent of maximum power); range 0% to 127%.
Currently actuator offboard control interferes with SITL lockstep.
Therefore, the least we can do is to warn a user and inform them how to
workaround the issue.
* PreFlightCheck: remove unused reportFailures flag
* Commander: pull all pre flight checks together on the PreFlightCheck class
* PreFlightCheck: separate checks into their own files
This fixes the case where the mixer_module would subscribe and use its
own test_motor publication which was created only to make sure the
topic is advertised and subsequent updates will work properly.
This happened in SITL lockstep because the timestamp would be 0 at the
very beginning, and hence elapsed time would be 0 as well.
This lead to an actuator publication which would then get lockstep out
of sync causing poll errors on the Gazebo side.
The plot of the attitude setpoint in the log did not show any values
because the message timestamp that the position control module sets
was overwritten by the PositionControl attitude generation.
by applying it directly to the attitude setpoint which is the output of
the position controller.
The problem was that before the input to the attitude setpoint generation
was adjusted to generate a level attitude with zero thrust keeping the
heading. I refactored the PositionControl class in #13262 to directly
generate the attitude setpoint output. So here I'm adjusting the attitude
setpoint to do the exact same thing as before but without interleaving
with the PositionControl logic.
getLockGuard relies on copy elision to work correctly, which the compiler
is not required to do (only with C++17).
If no copy elision happens, the mutex ends up being unlocked twice, and the
CS is executed with the mutex unlocked.
The patch also ensures that the same pattern cannot be used again.
This fixes the battery failsafe for the following corner cases:
- Battery failsafe set to Return but we can't do RTL because we don't
have a global position or home position. In this case we now switch to
Land. Land might end up in Descend in the failsafe state machine
later.
- Battery failsafe set to Land but we can't land because we don't have a
local position. In this case we switch to land anyway and then fall
back to descend in the failsafe state machine later.
The "fix" involves ignoring using the main_state_transition and
implementing the guards in place. This is a hack for now but should
cover the corner case until a more thorough refactor.
The different failsafe state machines have involved over time from
requirements and learnings based on developed solutions and products.
The implementations in various places will need to get consolidated in
the future.
Tested in SITL for Return and Land with and without GPS.
Daniel Agar
bazooka joe
xdwgood
PX4 BuildBot
Peter van der Perk
Beat Küng
LP
Jacob Dahl
Mike Wiatt
Tanja Baumann
Julian Oes
JaeyoungLim
Matthias Grob