The way I prefer to think of the I term is as an offset compensation. Because a feedback controller needs an error to produce a control signal, it will never reach a setpoint requiring control effort, the integral term is an attempt to learn the control effort required to reach a setpoint.
When I was 14, I started an apprenticeship school for industrial measurement and control (yes, we used to start young here) and there I learned about PID. As I never progressed in this career (I went to electronics and IT after that, today I'm a technical account manager at google) is interesting how it came back to haunt me in my hobby 😁
Great way to break down and communicate how I term works. I really needed this! In betaflight I term relax doesn’t run during fast throttle inputs right? Seems like some I term windup happens here too but maybe I term can’t be clamped due to throbbles 🤔
no it does not. for fast throttle moves, Anti-Gravity boosts P and I-Gains. If you have any wobble, you may need to increase the P ratio for that in the CLI. It boosting both is correct, but it may require more P in the P / I balance.
Nice video 👍🏻. I (excuse the pun) imagine the d term is not having much of an effect on the I term - because the oscillations are slow. D term responds to rate of change, not the quantity of change. Interesting enough, as the d term was increased - the oscillation became even slower, which would cause the d term to shrink. Is this the opposite to d term oscillation? 😂
How is the difference from external factors ( wind or any non set point input ) , between I and D ? Are they completely uncorrelated in this case too ? ( for a quadcopter )
how difficult would it be to have a simulator like above on actual blackbox data inputs and/or stick movements using the PID controller of betaflight? i do love the implementation in TRYP FPV and i get to play around alot with pid tuning based on your and chris rossers content. so far its been surprising how well they transfer to real life. i wonder about two things: - how closely can we model our actual quad in a simulated scenario and tune from there? - how difficult is it to build software into betaflight for the quad to self-tune based on blackbox analysis like setpoint response curves and noise vs filters, etc.?
Good questions! Sadly the answer is it will be difficult. The AI drone team went into the simulation vs real life mismatch, or cited sourc(es) which did. I expect you are almost entirely tuning to deal with horible fluidmechanical effects. All the other stuff you can probaply get reasonably close to ideal yourself. edit: so the zurich team apperently modeled their drone prety "accurately" (for what they where doing with it). But their publication wasn't about the feedback controller or filtering but about how to replace the human. The problems you encouter for what they did and what is required for automated controller and filter tuning is different and as far as I know automated tuning as you want it is already rare in very clean, high performance aplications.
The way I prefer to think of the I term is as an offset compensation. Because a feedback controller needs an error to produce a control signal, it will never reach a setpoint requiring control effort, the integral term is an attempt to learn the control effort required to reach a setpoint.
When I was 14, I started an apprenticeship school for industrial measurement and control (yes, we used to start young here) and there I learned about PID. As I never progressed in this career (I went to electronics and IT after that, today I'm a technical account manager at google) is interesting how it came back to haunt me in my hobby 😁
Host: "When Pid Pooning, a term that people don't really understand to well...."
Me: What the hell is Pid Pooning? I don't understand. :)
Great way to break down and communicate how I term works. I really needed this!
In betaflight I term relax doesn’t run during fast throttle inputs right? Seems like some I term windup happens here too but maybe I term can’t be clamped due to throbbles 🤔
no it does not.
for fast throttle moves, Anti-Gravity boosts P and I-Gains. If you have any wobble, you may need to increase the P ratio for that in the CLI. It boosting both is correct, but it may require more P in the P / I balance.
@@uavtechthanks! I’ll give that a try
Nice video 👍🏻. I (excuse the pun) imagine the d term is not having much of an effect on the I term - because the oscillations are slow. D term responds to rate of change, not the quantity of change.
Interesting enough, as the d term was increased - the oscillation became even slower, which would cause the d term to shrink. Is this the opposite to d term oscillation? 😂
basically, right. :-)
How is the difference from external factors ( wind or any non set point input ) , between I and D ? Are they completely uncorrelated in this case too ? ( for a quadcopter )
it works the same either way between I, P and D (external change or Setpoint change).
how difficult would it be to have a simulator like above on actual blackbox data inputs and/or stick movements using the PID controller of betaflight? i do love the implementation in TRYP FPV and i get to play around alot with pid tuning based on your and chris rossers content. so far its been surprising how well they transfer to real life.
i wonder about two things:
- how closely can we model our actual quad in a simulated scenario and tune from there?
- how difficult is it to build software into betaflight for the quad to self-tune based on blackbox analysis like setpoint response curves and noise vs filters, etc.?
Good questions! Sadly the answer is it will be difficult. The AI drone team went into the simulation vs real life mismatch, or cited sourc(es) which did. I expect you are almost entirely tuning to deal with horible fluidmechanical effects. All the other stuff you can probaply get reasonably close to ideal yourself.
edit: so the zurich team apperently modeled their drone prety "accurately" (for what they where doing with it). But their publication wasn't about the feedback controller or filtering but about how to replace the human. The problems you encouter for what they did and what is required for automated controller and filter tuning is different and as far as I know automated tuning as you want it is already rare in very clean, high performance aplications.
*following*
😜
Thank you for another great “layman” explanation of the PIDs.
Best wishes in 2024!
💯 quality content , graphs-animation makes it really easy to understand , well done . THX a lot 👍
Very interesting and this makes lot of sense! Still learning everyday 👍 Thanks for your great video, really helpful 🙏
Quality content. As always thanks sir
wow. .. this is great. am now getting a grip of what PID means
excellent
I>P>D