It is a calm day at the moment. I had a good auto tune on one axis. I changed the axis to yaw and lifted off again.
The drone went a little backwards and I pushed it a little forward. I then pushed it higher up so that I could get out of ground effect. It seemed ok.
I changed to loiter and that is when it tried to attack me. I was about 10m away but it came straight for me. I ran like a lion was chasing me and I tripped like in the movies. Lucky for me it hit the ground next to me.
Not sure why you changed the title. Doesn’t seem like correct English but if that is what works I am fine with it since English isn’t my first language.
I wonder why it picked those values with auto tune.
How would your manually tune? Would you also make the drone tilt in one direction and another? What then after the tilt? How do you know it was a good or bad tilt?
I read the manual tune procedure but I don’t know what is good or bad if I tune manually.
You won’t get a good Auto Tune if there is already a problem present. You can’t fly with those PID’s so I would set them back to default and set these to the Initial Tuning parameters values:
ATC_ACCEL_P_MAX
ATC_ACCEL_R_MAX
ATC_ACCEL_Y_MAX
Then do some manual tuning as per the guidelines. Using Transmitter Based Tuning takes less time. Also you have enabled thew Dynamic Notch filter but all but the throttle reference is at default. No point in that unless the frequency actually is 80Hz. So, either disable it or configure it properly.
A question about transmitter tuning: How do you manage the relationships between values when using the transmitter tuning? For example, rate P should equal rate I, but the transmitter tuning will only adjust one at a time (at least from what I can see in the wiki). If I’m tuning P, I know I can just set I value afterwards, but how will the drone react if I’m making big changes in P without moving I?
I have not found that to be a problem but you may have to return to another P tune session after setting I equal from the previous one. There is a lot of iteration which can be quite time consuming. I’ll usually run thru a complete P,I,D manual tune then increase the Angle pitch/roll P term and then go back to Rate P. At this point you are reviewing the logs often. I graph RCin6 and Pitch/Roll Actual/Desired and look for the best result in the RCin step. Give some aggressive Pitch/Roll, turn the tune knob a bit and repeat.
Tell me if this methodology makes sense to you: I was using transmitter tuning the other day to adjust D. (I felt that P/I were already acceptable, and also my earlier question…) I could see some oscillations in the logs around the Actual and I could hear the motors “burble”. Through pitch and roll inputs I couldn’t see anything in flight but I noticed that in a vertical descent (like in an RTL) I could get it to wobble in flight enough that I could see it. I did a series of climbs to ~20m and then gradually worked up to full speed descents, each time adjusting D until the quad came down nice and smooth. The final change was very slight from the previous D value but enough, so in future I would probably only use this for final fine tuning.
What do you think? Good approach or did I just get lucky?
I have done exactly that. I start with D but after tweaking all the other parameters I’ll sometimes have to came back to D. On my sub250 5" it was for the same reason you saw. I could hear it in the motors.
So do you guys lift up and go down and if there is a wobble you change the value?
The values at the top of the screen start at 4.5 I believe. Do you go up or down first?
You say you adjust the angle so do you also twitch like auto tune? What should happen when you twitch?
This is the text. What is overshoot? Is it that it tilts the opposite direction after you leave the stick?
If the aircraft begins to overshoot significantly or oscillate after the stick input, halt the tests before the situation begins to endanger the aircraft. The aircraft may require manual tuning (see next section) before autotune can be run.
I wasn’t sure the filter was working because the graph still showed a big spike at 80 after having the filter on.
The twitch inputs I’m putting in are a lot like what auto tune does. I start pretty small when I’m testing a tune just to make sure things don’t get out of hand and the inputs get larger as gain confidence the machine won’t go bad. Overshoot is when you’ve commanded a pitch/roll angle and it goes past that point. Probably easiest to see is when you have the quad moving in one direction (pitch or roll), and you quickly recenter the stick. Does the quad snap to level (good), or does it go past level and have to correct itself back (overshoot)? Or does it keep rocking back and forth as it settles into level (oscillation)?
You should be in alt-hold or even stabilized mode when you’re testing this.