Replies to This Discussion

Permalink Reply by Tom Yochum on October 1, 2010 at 9:04am

I highly recommend Aircraft Control and Simulation for aircraft control theory.

You have the right idea about stabilization. Generate an error signal (actual roll - desired roll, for example) and use a PID (or other algorithm) to drive a servo that makes the error go to zero.

• ▶ Reply

Permalink Reply by Simone Chiaretta on October 2, 2010 at 7:19am

Thank you, I'll see if I find a copy somewhere in a library, it's a bit pricely :)
Simo

• ▶ Reply

Permalink Reply by Tj Bordelon on October 1, 2010 at 2:19pm

PID loop is very simple. It outputs a value based on error. The first term is proportional to the error, the second is the integrated (over time) error value, and the last is the rate of change of the error.

Output = (error * P_GAIN) + (integrated_error * I_GAIN) + (error_rate_of_change * D_GAIN)

You can just start with the P stuff. Error is simply (command - actual), multiply that by a gain, and that's what feeds the output (servo, etc). That will settle at a value that is not quite right. That's where I comes in. It drives the DC (non-changing) error to zero.

Every "update" you integrate your error. So if you computed error like this:

error = command - actual

You can do the integral like this (using euler integration):

integrated_error = integrated_error + (error * dt)

... Where dt is the floating point timestep (ie, .1 for 1/10th of a second). This just happens over and over in the update loop.

The derivitave term is the same deal. That just "backs off" of the output the faster the error is changing. For example, if you are commanding a particular roll angle, and the plane starts to roll hard to correct, this term will grow larger and back off of the output a bit. It helps to prevent overshoot.

I usually just do P and PI loops, throwing in D when there's oscillations.

Here's an example to command a bank angle:

void update_roll_PID( float dt )
{
error = (roll_angle - desired_roll_angle);
integrated error += error * dt
aileron_servo_out = error * P_GAIN + integrated_error * I_GAIN + roll_gyro * D_GAIN;
}

So with these PID loops, I fly my plane with the following loops:
1) Desired heading PID ---> Bank angle PID --> ailerons
2) Altitude PID --> Pitch angle PID --> elevator
3) Airspeed PID --> throttle


So at this point you're commanding airspeed, heading, and desired altitude.

Hope this helps.

• ▶ Reply

Permalink Reply by Simone Chiaretta on October 2, 2010 at 7:25am

If with the PID you are controlling speed, heading and altitude you are controlling everything. What else is needed to fly something through a series of wayapoints?

• ▶ Reply

DeveloperPermalink Reply by Doug Weibel on October 2, 2010 at 8:51am

It is generally a little more complicated, but at the high level that is all that is needed. At a lower level, how do you control those things? We use cascaded PID control loops. For example a loop based on desired roll angle controls the aileron servos while a loop based on desired heading produces a desired roll angle.

• ▶ Reply

Permalink Reply by Steve on January 21, 2012 at 2:35am

Doug, since you are the autopilot jedi here. Can you please do a simple block diagram of a cascaded PID controller that you are talking about?

• ▶ Reply

Permalink Reply by Tj Bordelon on October 2, 2010 at 7:31pm

You add yet another layer above the lower level loops. "Fly to waypoint" code usually just computes the heading from where you are to where you're going and feeds the feedback loop I listed in #1: heading PID loop.

Then on top of that you can add even higher level logic.

• ▶ Reply

Permalink Reply by Simone Chiaretta on October 4, 2010 at 5:01am

Great... thank you very much... very great explanation
thank you
Simone

• ▶ Reply

Permalink Reply by Chris Kevorkian on October 3, 2010 at 7:33pm

this a great lil pdf that breaks down PID controllers a lil bit futher

http://igor.chudov.com/manuals/Servo-Tuning/PID-without-a-PhD.pdf

• ▶ Reply