Turn-to-angle PID tuning

[Oblarg]

So, we’re currently using a cascading PID control to implement a turn-to-angle function. That is, we’re running a position PID using the output of our gyro, whose outputs feed to our wheel velocity PID.

On the whole, it’s working fine, but there are some fine-tuning issues we’d like to address.

Currently, we are only using P.

Namely, if we make P high enough to get within 1-2 degrees of the setpoint in a reasonable amount of time, the robot builds up enough inertia while turning to overshoot the setpoint. Setting P just high enough to not overshoot the setpoint, we can get to ~5 degree accuracy reasonably quickly, but it takes quite a few seconds more to zero in.

There are two solutions I can think of for this - introducing a bit of I to provide some extra “push,” or setting a “minimum output” around 0 that small output values get raised to.

I’ve done the latter to overcome friction in an elevator control loop before, and it worked wonderfully. But my intuition is that it should not be needed for a cascading loop where the task of overcoming friction at small values should be handled by the inner velocity loop.

Any suggestions?

Addendum: We’re also using a similar P loop to keep the robot driving straight while a button is held, by adding the output as a difference term to our linear velocity. The output is scaled between -1 and 1, where 1 represents max robot speed. We were struggling to find a value of P that worked optimally - if anyone who has done similar things could share what values they have used in the past (with offset being measured in degrees), it’d be appreciated.

We would do further testing to answer these questions ourselves, but we have exhausted our 6 hour window to toy with the robot.

Thanks!

[lethc]

From my experience adding a ‘minimum’ output to your PID loop is by far the quickest, most painless solution. Our drivetrain has a large amount of friction and adding the minimum made it much more responsive and quicker to tune. It exponentially shortened PID tuning times for my team.

[Tom_Bottiglieri]

See this thread http://www.chiefdelphi.com/forums/showthread.php?t=145404

[Oblarg]

@Tom Bottiglieri

See this thread http://www.chiefdelphi.com/forums/showthread.php?t=145404

@Jared Russel

This is a constant source of annoyance for doing precise turning with high-friction wheels. Some solutions that have worked for us in the past:

1. Figure out what the minimum required output is to get the robot to turn at all. If the PID output is less than this magnitude, output this magnitude (with the same sign as the PID output).

2. Instead of using your heading PID loop to generate PWM commands directly, use it to generate velocity commands that a second PID loop on each side of the drive will attempt to follow. Note that good velocity control at low speeds requires a really fast control loop and accurate velocity measurement. The CAN Talon SRX and CTRE Mag Encoder have impressed us in this regard.

3. Motion profiling can help (a bit) to ensure that the robot smoothly glides to its final heading, but this method works best in conjunction with 1 and/or 2.

Well, this is quite nicely consistent with our observations. We’re already implementing suggestion 2) but we’re handling PID in the 50hz main loop which is probably unable to handle the low speeds adequately. Minimum output it is!

Thanks for the help, both (or all three?) of you.

If anyone has input for a good P value for the “drive straight” loop, though, it’d still be greatly appreciated.