Bench top setup for demonstrating PID

[apalrd]

A 'pure' PID controller can act as a velocity controller if both of the inputs (setpoint and process variable) are velocity. In this case, D will be unused, I will act as P, and P will act as D. The issues come when the PID class has additional logic to handle the I term, such as resetting the integral when the error zero-crosses, capping the integral, etc. - I've heard the Jaguars reset I under some conditions, although I've never verified this.

Every time I have worked with a speed controller, attempting to adapt a position controller has always failed to meet the design requirements, and I design a better controller to handle it (usually incorporating some application-specific limiting of the integral term). Adding a Feed Forward term also helped with response time and oscillation.

Basically, what I'm saying is, don't be afraid to write your own PID math and not use the per-existing class.

[Ether]

Originally Posted by apalrd

A 'pure' PID controller can act as a velocity controller if both of the inputs (setpoint and process variable) are velocity. In this case, D will be unused, I will act as P, and P will act as D. The issues come when the PID class has additional logic to handle the I term, such as resetting the integral when the error zero-crosses, capping the integral, etc. - I've heard the Jaguars reset I under some conditions, although I've never verified this.

Every time I have worked with a speed controller, attempting to adapt a position controller has always failed to meet the design requirements, and I design a better controller to handle it (usually incorporating some application-specific limiting of the integral term). Adding a Feed Forward term also helped with response time and oscillation.

Basically, what I'm saying is, don't be afraid to write your own PID math and not use the per-existing class.

Here are some other options:

http://www.chiefdelphi.com/forums/sh...90&postcount=5

[MrRoboSteve]

The Robodox paper proposes implementing a PIDSource subclass (RPMSource) that wraps an RPM value, which is the wired to the WPILib PIDController. On each turn of the Execute() crank in our command, we take the elapsed time and change in encoder count to calculate an RPM value, which is fed into the RPMSource instance.

apalard -- it does look like we're not too far from doing all of the calculations ourselves, so we will probably give that a try if the build-in controller fails us. One item that hasn't been obvious is how far back in time to accumulate errors.