Society of Robots - Robot Forum
Software => Software => Topic started by: quazimodo on October 19, 2008, 02:15:16 AM
-
So i am working on a bot and am attempting to implement some sort of PID control to match the speed of the wheels in differential drive.
Now, the wheels are diameter 35mm and speeds are between 0-60 RPM. Currently i have encoders on the wheels on either side with 18 state changes on each. My atmega88 polls for a state change and when it happens it increments a counter for each wheel along with the time since the last change in state.
The first approach i took was to run the PID loop each time a click happened and use the time since the last click to determine velocity. Unsuccessful.
the 2nd approach was to run the PID loop a number of times per second and just use that time between PID loops and the numbre of clicks accumulated to figure out the velocity (actually its not velocity, just speed as i dont actually know the direction from the encoders). Anyway both approaches have failed and i can not get this bloody thing to work right. When i want 1rpm the code will send a signal to the PWM that is too low to even move the bot. I think that i may have to create some sort of table because the motors react in a non linear way, ie duty cycle of 30% does not produce 30% of the speed of a 100% duty cycle, etc.
So basically im stuck... with 18 clicks per wheel encoder and speeds of under 60rpm how often should i be running this control loop? i am totally in the dark here, im not even sure how to program it.
as i have it now it goes like this for each wheel
poll encoder and if state changes increment counterA
every 0.1 seconds run PID function
PID function :
avgCounterA = average of current counterA and the previous 4 (as in the CounterA used in the previous 4 PID loops)
//the above line has an array and for loops, etc, to do it right... too long to write here
angular velocity = function of time between PID loops and AvgCounterA
error = wanted angular velocity vs angular velocity
output = oldoutput+ Kp * (error-olderror) + Ki * (error + old error) + Kd* (error - 2*olderror + 2ndolderror)
oldoutput = output
2ndolderror = olderror
olderror = error
counterA=0;
end
i have the average of counterA's because of the very low resolution of the wheel (am i right to do this)
Also i have not used the Integral or Differential code because i cant even get the proportional working right!
anyway, anyone who has ever done a slow robot with low resolution encoder PLEASE opine !!!
should i increase the encoder resolution or can i get away with 18 clicks?
how should i program this?
Oh a note, the range of values i can give the code that does the actual driving is the integer range 0-165, and the output is treated later to be within that range.
this may be too much to ask, but if someone has done something like this already, what is the chances of me seeing the code?
thanks so much
-
OK so nvm, i figured it out for the most part
Next time i do this im going to get my encoders fast enough or high density enough for atleast 2 clicks at the minimum desired speed (which will be quantised) with 0 clicks being zero velocity. A separate function can update velocity for the PID control loop to use later.
-
I have built a robot with servos (60 RPM) and I had 32 clicks per revolution. I calculated the encoders so they would give me 5mm traveled distance per click. I didn't had time to play too much with the PID on that robot since I had a hard time programming the micro (I was getting errors and I bricked it). My encoders were quadrature with a single channel interrupt and I think I was missing clicks at some point on the wheel rotation. So nothing was perfect.
I am building now a driving base and I want to mount the disk directly to the motor shaft with 14 clicks and I am using better optical sensors (Hamamatsu). I still have to find a way to mount the disk since the motor shaft is recessed. But anyway, this setup will give me 583 clicks per second or 364 clicks per wheel rotation. The DC motors have 100 RPM @ 6V and a 1:25 gear ratio.