Help with Steering?

[interfect]

Hello,
We have our robot set up with two independent steering wheels, tracked with encoders and run by globe motors geared down at about a 3:1 ratio. It has fallen to me to program this rig, and convince it to do something useful. I am trying to get the wheels to point in a certain direction, but I am having trouble with the position seeking algorithm.
Running the wheels right when they're turned too far left, and left when they are too far right, just makes them oscillate about the target (I didn't add a dead zone. I could try that, but I want them to seek exactly.)
I tried implementing full PID control, but tuning it has already consumed more time than it's really worth, and no matter what I do it either oscillates or doesn't get to the target, or sometimes both.
I think the underlying problem may be that, because of the motors and the gearing, the wheels don't perceptibly move unless the motor is set to more than about 20 units off of idle (127). Once the motor revs up, it's hard to stop on target. PID ought to be able to deal with this, but it isn't.
What sort of control schemes have other people used for this sort of problem? How do you tune your PID loops (if you use them)?

What's the simple & effective solution that I am ignoring?

[Alan Anderson]

I think PID is exactly the solution for your situation. You'll probably want to make the P constant as small as possible while still getting the motor to start moving reliably; that will help prevent oscillation. The D constant will keep the overshoot down, perhaps to the point that it won't reach the target. The I constant will account for that, and will likely be the most important to get right.

[MrForbes]

We set up a test Ackermann steering robot, and used a pot on one spindle, and a tie rod connecting both sides together. It uses a window motor and simple linkage to turn the wheels. It was connected to a vex controller, and programming it to run with a joystick for steering seemed to go pretty quickly (CD user TheOtherGuy did the programming), using a PID loop.



I dont know how the mechanical speed of this one compares to your setup....maybe that makes a difference in how sensitive the loop needs to be?

[Daniel_LaFleur]

Originally Posted by interfect

Hello,
We have our robot set up with two independent steering wheels, tracked with encoders and run by globe motors geared down at about a 3:1 ratio. It has fallen to me to program this rig, and convince it to do something useful. I am trying to get the wheels to point in a certain direction, but I am having trouble with the position seeking algorithm.
Running the wheels right when they're turned too far left, and left when they are too far right, just makes them oscillate about the target (I didn't add a dead zone. I could try that, but I want them to seek exactly.)
I tried implementing full PID control, but tuning it has already consumed more time than it's really worth, and no matter what I do it either oscillates or doesn't get to the target, or sometimes both.
I think the underlying problem may be that, because of the motors and the gearing, the wheels don't perceptibly move unless the motor is set to more than about 20 units off of idle (127). Once the motor revs up, it's hard to stop on target. PID ought to be able to deal with this, but it isn't.
What sort of control schemes have other people used for this sort of problem? How do you tune your PID loops (if you use them)?

What's the simple & effective solution that I am ignoring?

PID is really the way to go.

One of the things you'll need to do is slow down the turning motor as it gets closer to the target direction. If you dont do this you will end up hunting around your target (also caled classic occillation).

Also, if your wheels start to turn right when you are turning left then you are probably manipulating the input number such that it goes over 255 or below 0 and thus "wraps around". You need to check (Printf) what your numbers are when you are sending them, as a PWM, to the motor.

I would also suggest using a pot, rather than an encoder, to give you positional feedback. That should avoid 'wrapping around' with your PWM numbers.