Comments

• 

  Lorentz September 17, 2012 at 11:38am

  @ Jason

  that's reassuring. I indeed suspected that inaccuracies in the simulation model would have the most part .
• Developer

  Doug Weibel September 17, 2012 at 8:21am

  HI Lorenz,

  Jason's comments are (generally) correct.  With a non-linear system sometimes you get counter-intuitive and unusual results, hence the small disclaimer.  Regardless, adding delay into a feedback path almost always reduces stability margins and requires lower gains, and removing the delay will increase stability margins.

  To really choose appropriate gains through simulation is a very significant task, requiring really good models.  I just did some test flights for a new UAV autopilot startup called BlackSwift last week - they (two aerospace engineering PhD's) had gone to considerable effort to find some starting gains prior to actual flight. Other than being stable, they were not very good at all, and about an hour of actual flight time was required for tuning...  Unless you have a really good physics model and really good airframe model I think that is about all you can get through simulation - conservative gains that at least you can expect to be stable.
• 

  Strike Eagle III September 17, 2012 at 7:28am

  @Lorenz - I think there are bigger issues with using HIL simulation for gain tuning such as imperfections in the physics / aerodynamics models employed by X-plane and imperfections in the model of your airplane in X-plane.  They are probably good enough for a starting point, but I think the only way to get good polished gains is from real flight.

  To answer your question though, delays have a destabilizing effect, so if I had to guess, I would say any gains that work with the delays in place would work better (and be a little conservative) when those delays are removed.
• 

  Lorentz September 17, 2012 at 5:02am

  
  I don't know if this is the correct place to ask a aquestion regarding HIL simulation.

  I was trying to understand the effects of delays on the tuning of PID parameters.

  Has anybody been able to measure the delay between movements / disturbances of the aircraft in X-Plane and the availability of relevant data (attitude, sensors, ...) to the ArduPilot code ?

  This includes X-Plane sending data to ArduPilotMegaPlanner, that elaborates them and sends to ArduPilotMega board via serial communication.

  Since those delays are simply not there in the real flight situation, I suspect that the PID parameters resulting from HIL simulation tuning will be different from the ones for the real situation.

  Sincerely, I'd be happy if anybody could confute this.

  Regards
• 

  Jesper Rooth September 12, 2012 at 2:05pm

  OK. Tried that already since I always clear logs after each flight. But problem seems to go away when I downgrade to 2.50. Then all logs get in the right order and can be downloaded. Hmmm..
• 3D Robotics

  Chris Anderson September 12, 2012 at 1:29pm

  Jesper: No, datalogging is a different chip. To reset/erase that, do a logs/erase
• 

  Jesper Rooth September 12, 2012 at 1:22pm

  Would CLI setup/erase/reset also reset the datalogging memory? I have wierd logging now with logs have higher start numbers than stop.
• 3D Robotics

  Chris Anderson September 12, 2012 at 12:31pm

  velentin: probably some got corrupted in the EEPROM. Do a CLI setup/erase /reset and go through the config process again and you should be fine. 
• 

  valentin colon September 12, 2012 at 12:20pm

  hi there, I'm having problem with APM2, the response of the sensor is extremely low this is the second time that happens, I solved it re installing the codes I would like to find out what is causing this
• 

  Brandon Jones July 25, 2012 at 7:25am

  @Tomas:  The best way is to directly look at the code, attitude.pde and navigation.pde.  It is commented pretty well.  From the wiki just click on the source tab and browse...