MOtor control Pololu
Dependencies: Encoder HIDScope mbed-dsp mbed
main.cpp
- Committer:
- vsluiter
- Date:
- 2014-09-29
- Revision:
- 5:06381e54b94a
- Parent:
- 4:1a53b06eeb7f
- Child:
- 6:0832c6c6bcba
File content as of revision 5:06381e54b94a:
#include "mbed.h" #include "encoder.h" #include "MODSERIAL.h" #define TSAMP 0.01 #define K_P (10 *TSAMP) #define K_I (0.00 *TSAMP) #define K_D (5 *TSAMP) #define POT_AVG 50 void coerce(float * in, float min, float max); float pid(float setpoint, float measurement); AnalogIn potmeter(PTC2); volatile bool looptimerflag; float potsamples[POT_AVG]; void setlooptimerflag(void) { looptimerflag = true; } void potAverager(void) { static uint16_t sample_index = 0; float voltage = potmeter.read()-.5; potsamples[sample_index] = voltage; sample_index++; if(sample_index >= POT_AVG) sample_index = 0; } float getpotAverage(void) { uint16_t valuecounter; float sum = 0; for(valuecounter = 0 ; valuecounter < POT_AVG ; valuecounter++) { sum += potsamples[valuecounter]; } return sum / (POT_AVG*1.); } int main() { Encoder motor1(PTD0,PTC9); Ticker potaverage; MODSERIAL pc(USBTX,USBRX); PwmOut pwm_motor(PTA12); DigitalOut motordir(PTD3); potaverage.attach(potAverager,0.002); pc.baud(921600); Ticker looptimer; looptimer.attach(setlooptimerflag,TSAMP); pc.printf("bla"); while(1) { float setpoint; float new_pwm; while(!looptimerflag); looptimerflag = false; setpoint = (getpotAverage())*2000; //new_pwm = (setpoint - motor1.getPosition())*.001; new_pwm = pid(setpoint, motor1.getPosition()); coerce(&new_pwm, -1,1); pc.printf("%f %f %f 0\n", setpoint/2000.+0.5, motor1.getPosition()/2000.+.5,new_pwm/2.+0.5); if(new_pwm > 0) motordir = 1; else motordir = 0; pwm_motor.write(abs(new_pwm)); } } //coerces value 'in' to min or max when exceeding those values //if you'd like to understand the statement below take a google for //'ternary operators'. void coerce(float * in, float min, float max) { *in > min ? *in < max? : *in = max: *in = min; } float pid(float setpoint, float measurement) { float error; static float prev_error = 0; float out_p = 0; static float out_i = 0; float out_d = 0; error = setpoint-measurement; out_p = error*K_P; out_i += error*K_I; out_d = (error-prev_error)*K_D; coerce(&out_i,-0.5,0.5); prev_error = error; return out_p + out_i + out_d; }