File content as of revision 17:6263e90bf3ba:

//***************************************************************************************
//Kalman Filter implementation
//***************************************************************************************
#include "Kalman.h"
#include "rtos.h"
#include "math.h"
#include "supportfuncs.h"
//#include "globals.h"

#include <tvmet/Matrix.h>
using namespace tvmet;



namespace Kalman
{

//State variables
Vector<float, 3> X;
Matrix<float, 3, 3> P;
Mutex statelock;

float RawReadings[maxmeasure+1];
float SensorOffsets[maxmeasure+1] = {0};

bool Kalman_init = 0;

struct measurmentdata {
    measurement_t mtype;
    float value;
    float variance;
}

Mail <measurmentdata, 16> measureMQ;



//Note: this init function assumes that the robot faces east, theta=0, in the +x direction
void KalmanInit()
{

    //Solving for sonar bias is done by entering the following into wolfram alpha
    //(a-f)^2 = x^2 + y^2; (b-f)^2 = (x-3)^2 + y^2; (c-f)^2 = (x-1.5)^2+(y-2)^2: solve for x,y,f
    //where a, b, c are the measured distances, and f is the bias

    SensorOffsets[SONAR0] = sonartimebias;
    SensorOffsets[SONAR1] = sonartimebias;
    SensorOffsets[SONAR2] = sonartimebias;

    //solve for our position (assume perfect bias)
    const float d = beaconpos[0].y - beaconpos[1].y;
    const float i = beaconpos[0].y - beaconpos[2].y;
    const float j = beaconpos[0].x - beaconpos[2].x;

    float y_coor = (r1*r1-r2*r2+d*d)/2d;
    float x_coor = (r1*r1-r3*r3+i*i+j*j)/(2*j) - (i*y_coor)/j;

    //IR

    float IRMeasuresloc[3];
    IRMeasuresloc[0] = RawReadings[IR0];
    IRMeasuresloc[1] = RawReadings[IR1];
    IRMeasuresloc[2] = RawReadings[IR2];
    //printf("0: %0.4f, 1: %0.4f, 2: %0.4f \n\r", IRMeasuresloc[0]*180/PI, IRMeasuresloc[1]*180/PI, IRMeasuresloc[2]*180/PI);

    float IR_Offsets[3];
    float fromb0offset = 0;
    for (int i = 0; i < 3; i++) {

        //Compute IR offset
        float angle_est = atan2(beaconpos[i].y - y_coor,beaconpos[i].x - x_coor);

        //printf("Angle %d : %f \n\r",i,angle_est*180/PI );
        IR_Offset[i] = constrainAngle(IRMeasuresloc[i] - angle_est);
        
        fromb0offset += IR_Offsets[i] - IR_Offset[0];
    }


    //debug
    printf("Offsets IR: %0.4f, Sonar: %0.4f \r\n",IR_Offset*180/PI,Sonar_Offset*1000 );


    statelock.lock();
    X(0) = x_coor/1000.0f;
    X(1) = y_coor/1000.0f;

    if (Colour)
        X(2) = 0;
    else
        X(2) = PI;
    statelock.unlock();


    //reattach the IR processing
    ir.attachisr();
}


void Kalman::predictloop(void* dummy)
{

    OLED4 = !ui.regid(0, 3);
    OLED4 = !ui.regid(1, 4);

    float lastleft = 0;
    float lastright = 0;

    while (1) {
        Thread::signal_wait(0x1);
        OLED1 = !OLED1;

        int leftenc = encoders.getEncoder1();
        int rightenc = encoders.getEncoder2();

        float dleft = encoders.encoderToDistance(leftenc-lastleft)/1000.0f;
        float dright = encoders.encoderToDistance(rightenc-lastright)/1000.0f;

        lastleft = leftenc;
        lastright = rightenc;


        //The below calculation are in body frame (where +x is forward)
        float dxp, dyp,d,r;
        float thetap = (dright - dleft)*PI / (float(robotCircumference)/1000.0f);
        if (abs(thetap) < 0.02) { //if the rotation through the integration step is small, approximate with a straight line to avoid numerical error
            d = (dright + dleft)/2.0f;
            dxp = d*cos(thetap/2.0f);
            dyp = d*sin(thetap/2.0f);

        } else { //calculate circle arc
            //float r = (right + left) / (4.0f * PI * thetap);
            r = (dright + dleft) / (2.0f*thetap);
            dxp = abs(r)*sin(thetap);
            dyp = r - r*cos(thetap);
        }

        statelock.lock();

        float tempX2 = X(2);
        //rotating to cartesian frame and updating state
        X(0) += dxp * cos(X(2)) - dyp * sin(X(2));
        X(1) += dxp * sin(X(2)) + dyp * cos(X(2));
        X(2) = rectifyAng(X(2) + thetap);

        //Linearising F around X
        float avgX2 = (X(2) + tempX2)/2.0f;
        Matrix<float, 3, 3> F;
        F = 1, 0, (dxp * -sin(avgX2) - dyp * cos(avgX2)),
        0, 1, (dxp * cos(avgX2) - dyp * sin(avgX2)),
        0, 0, 1;

        //Generating forward and rotational variance
        float varfwd = fwdvarperunit * abs(dright + dleft) / 2.0f;
        float varang = varperang * abs(thetap);
        float varxydt = xyvarpertime * PREDICTPERIOD/1000.0f;
        float varangdt = angvarpertime * PREDICTPERIOD/1000.0f;

        //Rotating into cartesian frame
        Matrix<float, 2, 2> Qsub,Qsubrot,Qrot;
        Qsub = varfwd + varxydt, 0,
        0, varxydt;

        Qrot = Rotmatrix(X(2));

        Qsubrot = Qrot * Qsub * trans(Qrot);

        //Generate Q
        Matrix<float, 3, 3> Q;//(Qsubrot);
        Q = Qsubrot(0,0), Qsubrot(0,1), 0,
        Qsubrot(1,0), Qsubrot(1,1), 0,
        0, 0, varang + varangdt;

        P = F * P * trans(F) + Q;

        //Update UI
        float statecpy[] = {X(0), X(1), X(2)};
        ui.updateval(0, statecpy, 3);

        float Pcpy[] = {P(0,0), P(0,1), P(1,0), P(1,1)};
        ui.updateval(1, Pcpy, 4);

        statelock.unlock();
    }
}

void Kalman::runupdate(measurement_t type, float value, float variance)
{
    if (!Kalman_init)
        RawReadings[type] = value;
    else {

        RawReadings[type] = value - SensorOffsets[type];

        measurmentdata* measured = (measurmentdata*)measureMQ.alloc();
        if (measured) {
            measured->mtype = type;
            measured->value = value;
            measured->variance = variance;

            osStatus putret = measureMQ.put(measured);
            if (putret)
                OLED4 = 1;
            //    printf("putting in MQ error code %#x\r\n", putret);
        } else {
            OLED4 = 1;
            //printf("MQalloc returned NULL ptr\r\n");
        }

    }

}

void Kalman::updateloop(void* dummy)
{

    //sonar Y chanels
    ui.regid(2, 1);
    ui.regid(3, 1);
    ui.regid(4, 1);

    //IR Y chanels
    ui.regid(5, 1);
    ui.regid(6, 1);
    ui.regid(7, 1);

    measurement_t type;
    float value,variance,rbx,rby,expecdist,Y;
    float dhdx,dhdy;
    bool aborton2stddev = false;

    Matrix<float, 1, 3> H;

    float S;
    Matrix<float, 3, 3> I3( identity< Matrix<float, 3, 3> >() );


    while (1) {
        OLED2 = !OLED2;

        osEvent evt = measureMQ.get();

        if (evt.status == osEventMail) {

            measurmentdata &measured = *(measurmentdata*)evt.value.p;
            type = measured.mtype; //Note, may support more measurment types than sonar in the future!
            value = measured.value;
            variance = measured.variance;

            // don't forget to free the memory
            measureMQ.free(&measured);

            if (type <= maxmeasure) {

                if (type <= SONAR3) {

                    InitLock.lock();
                    float dist = value / 1000.0f - Sonar_Offset; //converting to m from mm,subtract the offset
                    InitLock.unlock();

                    int sonarid = type;
                    aborton2stddev = true;

                    statelock.lock();
                    //update the current sonar readings
                    SonarMeasures[sonarid] = dist;

                    rbx = X(0) - beaconpos[sonarid].x/1000.0f;
                    rby = X(1) - beaconpos[sonarid].y/1000.0f;

                    expecdist = hypot(rbx, rby);//sqrt(rbx*rbx + rby*rby);
                    Y = dist - expecdist;

                    //send to ui
                    ui.updateval(sonarid+2, Y);

                    dhdx = rbx / expecdist;
                    dhdy = rby / expecdist;

                    H = dhdx, dhdy, 0;

                } else if (type <= IR3) {

                    aborton2stddev = false;
                    int IRidx = type-3;

                    // subtract the IR offset
                    InitLock.lock();
                    value -= IR_Offset;
                    InitLock.unlock();

                    statelock.lock();
                    IRMeasures[IRidx] = value;

                    rbx = X(0) - beaconpos[IRidx].x/1000.0f;
                    rby = X(1) - beaconpos[IRidx].y/1000.0f;

                    float expecang = atan2(-rby, -rbx) - X(2);
                    Y = rectifyAng(value - expecang);

                    //send to ui
                    ui.updateval(IRidx + 5, Y);

                    float dstsq = rbx*rbx + rby*rby;
                    H = -rby/dstsq, rbx/dstsq, -1;
                }

                Matrix<float, 3, 1> PH (P * trans(H));
                S = (H * PH)(0,0) + variance;

                if (aborton2stddev && Y*Y > 4 * S) {
                    statelock.unlock();
                    continue;
                }

                Matrix<float, 3, 1> K (PH * (1/S));

                //Updating state
                X += col(K, 0) * Y;
                X(2) = rectifyAng(X(2));

                P = (I3 - K * H) * P;

                statelock.unlock();

            }

        } else {
            OLED4 = 1;
            //printf("ERROR: in updateloop, code %#x", evt);
        }

    }

}

// reset kalman states
void Kalman::KalmanReset()
{
    float SonarMeasuresx1000[3];
    statelock.lock();
    SonarMeasuresx1000[0] = SonarMeasures[0]*1000.0f;
    SonarMeasuresx1000[1] = SonarMeasures[1]*1000.0f;
    SonarMeasuresx1000[2] = SonarMeasures[2]*1000.0f;
    //printf("0: %0.4f, 1: %0.4f, 2: %0.4f \n\r", IRMeasuresloc[0]*180/PI, IRMeasuresloc[1]*180/PI, IRMeasuresloc[2]*180/PI);

    float d = beaconpos[2].y - beaconpos[1].y;
    float i = beaconpos[0].y - beaconpos[1].y;
    float j = beaconpos[0].x - beaconpos[1].x;
    float origin_x = beaconpos[1].x;
    float y_coor = (SonarMeasuresx1000[1]*SonarMeasuresx1000[1]- SonarMeasuresx1000[2]*SonarMeasuresx1000[2] + d*d) / (2*d);
    float x_coor = origin_x +(SonarMeasuresx1000[1]*SonarMeasuresx1000[1] - SonarMeasuresx1000[0]*SonarMeasuresx1000[0] + i*i + j*j)/(2*j) - i*y_coor/j;

    //statelock already locked
    X(0) = x_coor/1000.0f;
    X(1) = y_coor/1000.0f;



    /*    if (Colour){
            X(0) = 0.2;
            X(1) = 0.2;
            //X(2) = 0;
            }
        else {
            X(0) = 2.8;
            X(1) = 0.2;
            //X(2) = PI;
        }
        */
    P = 0.05, 0, 0,
    0, 0.05, 0,
    0, 0, 0.04;

    // unlocks mutexes
    statelock.unlock();

}

} //Kalman Namespace