The big news in this release is the addition of the new L1 navigation controller. The work on the L1 controller library was started by Brandon Jones, and his initial github pull request is a great please to look to understand the new code. As is so often the case with ArduPlane, the documentation still hasn't caught up with the code (though I hope it soon will!)
After Brandon’s initial work Paul Riseborough did a lot of work to add more robustness to the algorithm. The base L1 algorithm suffered from from instability in "path capture", which is how it handles navigation when you are a long way off course (for example, when you enter auto after having flown manually). Paul made some fantastic improvements which have been extensively flight tested by Brandon, Paul, Jon and myself.
I did a final test flight of it this morning in a Bixler2, and it flew very nicely. I found a setting of NAVL1_PERIOD=20 suited the Bixler2 very well. It produced great path tracking, and very nice circles in loiter, despite high wind.
Paul is not finished yet though! He's been working on re-working the APM_Control library started by Jon Challinger, and is getting fantastic results with roll and pitch control, greatly reducing the altitude loss in turns. That isn't in the 2.72 release though as I want to do a bit more testing and tweaking of the code, but I expect to get Paul’s new controllers in the next release.
Before I go into detail on how to tune the new L1 controller, I better list some of the other changes in this release:
- avoid doing auto radio trim if the resulting trim leaves less than 20% of travel between the trim value and the MAX and MIN for the aileron, elevator or rudder. This prevents auto-trim leaving you with very bad trim if you do it while your transmitter is off
- removed HIL_MODE_ATTITUDE, we always now use SENSORS HIL, which is much more accurate, and is needed for correct operation of the L1 controller
- added new LEVEL_ROLL_LIMIT parameter. This replaces both the TKOFF_HEAD_HOLD and RUDDER_STEER options, and instead allows you to specify a roll limit in degrees for auto land and takeoff. The default is 5 degrees, which should be small enough to prevent wings scraping on the runway in most planes.
- Removed the AHRS_BARO_USE option, as it caused trouble for too many users
- allow the WP_LOITER_RAD loiter radius to be set negative to mean that a counter clockwise loiter should be performed by default
- fixed the handling of the LOITER_TURNS and LOITER_TIME commands
- added new dataflash logging system, with self-describing logging format. See notes below for more info.
- on auto-takeoff, if we have an airspeed sensor then keep throttle suppressed until we have both 5m/s ground speed and 5m/s airspeed
- added support for 12 output PWM channels on PX4
- enable relay in camera library for all APM boards (thanks to SSzilard)
- cleanup use of const in libraries (thanks to Tobias)
- lowered default AHRS roll-pitch correction gains from 0.4 to 0.3
- changed AHRS to lower roll-pitch gains by another 50% when a high X acceleration is detected for a catapult launch, to prevent nose down on fast launch due to GPS velocity lag
- fixed RC throttle failsafe handling on PX4
- fixed a bug in returning airspeed estimate from DCM if an airspeed sensor is available (thanks to Jon Challinger for finding this important bug)
A few special notes about the DataFlash logging changes. This is the first release to support 'dataflash' logging on the PX4, and it implements it by writing binary log files to the SD card in an APM/logs sub-directory. Those log files are self-describing, which means the format of the logs is contained in the headers of the log. This makes it possible to add new log messages without having to change MissionPlanner. It also means you can now get logs into MissionPlanner by pulling the SD card out of the PX4 and plugging them into your PC, which is much faster than using the CLI. The new log format is also used for APM1 and APM2, which will mean we no longer have to align the version of MissionPlanner with the ArduPlane log version when analyzing dataflash logs.
Now on to the main thing in this release - the new L1 navigation controller.
The new controller makes a huge difference to navigation in ArduPlane in the following ways:
- tuning is now much simpler - the only navigation parameter you really have to change is NAVL1_PERIOD
- much better path tracking, even in high wind
- loiter now produces a very nice circle even in high wind
- much less servo flapping! If you tried to tune the old controller for very good path tracking it would end up changing servo position very rapidly, wearing out the servos and drawing more current than it should. The new controller has much smoother servo movement.
I hope you all enjoy flying this release as much as I do. Happy flying!