Prof Greg Egan
/
UAVXArm-GKE
UAVX Multicopter Flight Controller.
attitude.c
- Committer:
- gke
- Date:
- 2011-04-26
- Revision:
- 2:90292f8bd179
- Parent:
- 1:1e3318a30ddd
File content as of revision 2:90292f8bd179:
// =============================================================================================== // = UAVXArm Quadrocopter Controller = // = Copyright (c) 2008 by Prof. Greg Egan = // = Original V3.15 Copyright (c) 2007 Ing. Wolfgang Mahringer = // = http://code.google.com/p/uavp-mods/ = // =============================================================================================== // This is part of UAVXArm. // UAVXArm is free software: you can redistribute it and/or modify it under the terms of the GNU // General Public License as published by the Free Software Foundation, either version 3 of the // License, or (at your option) any later version. // UAVXArm is distributed in the hope that it will be useful,but WITHOUT ANY WARRANTY; without // even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. // See the GNU General Public License for more details. // You should have received a copy of the GNU General Public License along with this program. // If not, see http://www.gnu.org/licenses/ #include "UAVXArm.h" // Reference frame is positive X forward, Y left, Z down, Roll right, Pitch up, Yaw CW. // CAUTION: Because of the coordinate frame the LR Acc sense must be negated for roll compensation. void AdaptiveYawLPFreq(void); void DoLegacyYawComp(uint8); void NormaliseAccelerations(void); void AttitudeTest(void); void InitAttitude(void); real32 AccMagnitude; real32 EstAngle[3][MaxAttitudeScheme]; real32 EstRate[3][MaxAttitudeScheme]; real32 Correction[2]; real32 YawFilterLPFreq; real32 dT, dTOn2, dTR, dTmS; uint32 uSp; uint8 AttitudeMethod = Wolferl; //Integrator, Wolferl MadgwickIMU PremerlaniDCM MadgwickAHRS, MultiWii; void AdaptiveYawLPFreq(void) { // Filter LP freq is decreased with reduced yaw stick deflection YawFilterLPFreq = ( MAX_YAW_FREQ*abs(DesiredYaw) / RC_NEUTRAL ); YawFilterLPFreq = Limit(YawFilterLPFreq, 0.5, MAX_YAW_FREQ); } // AdaptiveYawFilterA real32 HE; void DoLegacyYawComp(uint8 S) { #define COMPASS_MIDDLE 10 // yaw stick neutral dead zone #define DRIFT_COMP_YAW_RATE QUARTERPI // Radians/Sec static int16 Temp; // Yaw Angle here is meant to be interpreted as the Heading Error Rate[Yaw] = Gyro[Yaw]; Temp = DesiredYaw - Trim[Yaw]; if ( F.CompassValid ) // CW+ if ( abs(Temp) > COMPASS_MIDDLE ) { DesiredHeading = Heading; // acquire new heading Angle[Yaw] = 0.0; } else { HE = MinimumTurn(DesiredHeading - Heading); HE = Limit1(HE, SIXTHPI); // 30 deg limit HE = HE*K[CompassKp]; Rate[Yaw] = -Limit1(HE, DRIFT_COMP_YAW_RATE); } else { DesiredHeading = Heading; Angle[Yaw] = 0.0; } Angle[Yaw] += Rate[Yaw]*dT; // Angle[Yaw] = Limit1(Angle[Yaw], K[YawIntLimit]); } // DoLegacyYawComp void NormaliseAccelerations(void) { const real32 MIN_ACC_MAGNITUDE = 0.7; // below this the accelerometers are deemed unreliable - falling? static real32 ReNorm; AccMagnitude = sqrt(Sqr(Acc[BF]) + Sqr(Acc[LR]) + Sqr(Acc[UD])); F.AccMagnitudeOK = AccMagnitude > MIN_ACC_MAGNITUDE; if ( F.AccMagnitudeOK ) { ReNorm = 1.0 / AccMagnitude; Acc[BF] *= ReNorm; Acc[LR] *= ReNorm; Acc[UD] *= ReNorm; } else { Acc[LR] = Acc[BF] = 0.0; Acc[UD] = 1.0; } } // NormaliseAccelerations void GetAttitude(void) { static uint32 Now; static uint8 i; if ( GyroType == IRSensors ) GetIRAttitude(); else { GetGyroRates(); GetAccelerations(); } Now = uSClock(); dT = ( Now - uSp)*0.000001; dTOn2 = 0.5 * dT; dTR = 1.0 / dT; uSp = Now; GetHeading(); // only updated every 50mS but read continuously anyway if ( GyroType == IRSensors ) { for ( i = 0; i < (uint8)2; i++ ) { Rate[i] = ( Angle[i] - Anglep[i] )*dT; Anglep[i] = Angle[i]; } Rate[Yaw] = 0.0; // need Yaw gyro! } else { DebugPin = true; NormaliseAccelerations(); // rudimentary check for free fall etc // Wolferl DoWolferl(); #ifdef INC_ALL_SCHEMES // Complementary DoCF(); // Kalman DoKalman(); //Premerlani DCM DoDCM(); // MultiWii DoMultiWii(); // Madgwick IMU // DoMadgwickIMU(Gyro[Roll], Gyro[Pitch], Gyro[Yaw], Acc[BF], -Acc[LR], -Acc[UD]); //#define INC_IMU2 #ifdef INC_IMU2 DoMadgwickIMU2(Gyro[Roll], Gyro[Pitch], Gyro[Yaw], Acc[BF], -Acc[LR], -Acc[UD]); #else Madgwick IMU April 30, 2010 Paper Version #endif // Madgwick AHRS BROKEN DoMadgwickAHRS(Gyro[Roll], Gyro[Pitch], Gyro[Yaw], Acc[BF], -Acc[LR], -Acc[UD], Mag[BF].V, Mag[LR].V, -Mag[UD].V); // Integrator - REFERENCE/FALLBACK DoIntegrator(); #endif // INC_ALL_SCHEMES Angle[Roll] = EstAngle[Roll][AttitudeMethod]; Angle[Pitch] = EstAngle[Pitch][AttitudeMethod]; DebugPin = false; } F.NearLevel = Max(fabs(Angle[Roll]), fabs(Angle[Pitch])) < NAV_RTH_LOCKOUT; } // GetAttitude //____________________________________________________________________________________________ // Integrator void DoIntegrator(void) { static uint8 g; for ( g = 0; g < (uint8)3; g++ ) { EstRate[g][Integrator] = Gyro[g]; EstAngle[g][Integrator] += EstRate[g][Integrator]*dT; // EstAngle[g][Integrator] = DecayX(EstAngle[g][Integrator], 0.0001*dT); } } // DoIntegrator //____________________________________________________________________________________________ // Original simple accelerometer compensation of gyros developed for UAVP by Wolfgang Mahringer // and adapted for UAVXArm void DoWolferl(void) { // NO YAW ESTIMATE const real32 WKp = 0.13; // 0.1 static real32 Grav[2], Dyn[2]; static real32 CompStep; Rate[Roll] = Gyro[Roll]; Rate[Pitch] = Gyro[Pitch]; if ( F.AccMagnitudeOK ) { CompStep = WKp*dT; // Roll Grav[LR] = -sin(EstAngle[Roll][Wolferl]); // original used approximation for small angles Dyn[LR] = 0.0; //Rate[Roll]; // lateral acceleration due to rate - do later:). Correction[LR] = -Acc[LR] + Grav[LR] + Dyn[LR]; // Acc is reversed Correction[LR] = Limit1(Correction[LR], CompStep); EstAngle[Roll][Wolferl] += Rate[Roll]*dT; EstAngle[Roll][Wolferl] += Correction[LR]; // Pitch Grav[BF] = -sin(EstAngle[Pitch][Wolferl]); Dyn[BF] = 0.0; // Rate[Pitch]; Correction[BF] = Acc[BF] + Grav[BF] + Dyn[BF]; Correction[BF] = Limit1(Correction[BF], CompStep); EstAngle[Pitch][Wolferl] += Rate[Pitch]*dT; EstAngle[Pitch][Wolferl] += Correction[BF]; } else { EstAngle[Roll][Wolferl] += Rate[Roll]*dT; EstAngle[Pitch][Wolferl] += Rate[Pitch]*dT; } } // DoWolferl //_________________________________________________________________________________ // Complementary Filter originally authored by RoyLB // http://www.rcgroups.com/forums/showpost.php?p=12082524&postcount=1286 const real32 TauCF = 1.1; real32 AngleCF[2] = {0,0}; real32 F0[2] = {0,0}; real32 F1[2] = {0,0}; real32 F2[2] = {0,0}; real32 CF(uint8 a, real32 NewAngle, real32 NewRate) { if ( F.AccMagnitudeOK ) { F0[a] = (NewAngle - AngleCF[a])*Sqr(TauCF); F2[a] += F0[a]*dT; F1[a] = F2[a] + (NewAngle - AngleCF[a])*2.0*TauCF + NewRate; AngleCF[a] = (F1[a]*dT) + AngleCF[a]; } else AngleCF[a] += NewRate*dT; return ( AngleCF[a] ); // This is actually the current angle, but is stored for the next iteration } // CF void DoCF(void) { // NO YAW ANGLE ESTIMATE EstAngle[Roll][Complementary] = CF(Roll, asin(-Acc[LR]), Gyro[Roll]); EstAngle[Pitch][Complementary] = CF(Pitch, asin(-Acc[BF]), Gyro[Pitch]); // zzz minus??? EstRate[Roll][Complementary] = Gyro[Roll]; EstRate[Pitch][Complementary] = Gyro[Pitch]; } // DoCF //____________________________________________________________________________________________ // The DCM formulation used here is due to W. Premerlani and P. Bizard in a paper entitled: // Direction Cosine Matrix IMU: Theory, Draft 17 June 2009. This paper draws upon the original // work by R. Mahony et al. - Thanks Rob! // SEEMS TO BE A FAIRLY LARGE PHASE DELAY OF 2 SAMPLE INTERVALS void DCMNormalise(void); void DCMDriftCorrection(void); void DCMMotionCompensation(void); void DCMUpdate(void); void DCMEulerAngles(void); real32 RollPitchError[3] = {0,0,0}; real32 OmegaV[3] = {0,0,0}; // corrected gyro data real32 OmegaP[3] = {0,0,0}; // proportional correction real32 OmegaI[3] = {0,0,0}; // integral correction real32 Omega[3] = {0,0,0}; real32 DCM[3][3] = {{1,0,0 },{0,1,0} ,{0,0,1}}; real32 U[3][3] = {{0,1,2},{ 3,4,5} ,{6,7,8}}; real32 TempM[3][3] = {{0,0,0},{0,0,0},{ 0,0,0}}; real32 AccV[3]; void DCMNormalise(void) { static real32 Error = 0; static real32 Renorm = 0.0; static boolean Problem; static uint8 r; Error = -VDot(&DCM[0][0], &DCM[1][0])*0.5; //eq.19 VScale(&TempM[0][0], &DCM[1][0], Error); //eq.19 VScale(&TempM[1][0], &DCM[0][0], Error); //eq.19 VAdd(&TempM[0][0], &TempM[0][0], &DCM[0][0]); //eq.19 VAdd(&TempM[1][0], &TempM[1][0], &DCM[1][0]); //eq.19 VCross(&TempM[2][0],&TempM[0][0], &TempM[1][0]); // c= a*b eq.20 Problem = false; for ( r = 0; r < (uint8)3; r++ ) { Renorm = VDot(&TempM[r][0], &TempM[r][0]); if ( (Renorm < 1.5625) && (Renorm > 0.64) ) Renorm = 0.5*(3.0 - Renorm); //eq.21 else if ( (Renorm < 100.0) && (Renorm > 0.01) ) Renorm = 1.0 / sqrt( Renorm ); else Problem = true; VScale(&DCM[r][0], &TempM[r][0], Renorm); } if ( Problem ) { // Divergent - force to initial conditions and hope! DCM[0][0] = 1.0; DCM[0][1] = 0.0; DCM[0][2] = 0.0; DCM[1][0] = 0.0; DCM[1][1] = 1.0; DCM[1][2] = 0.0; DCM[2][0] = 0.0; DCM[2][1] = 0.0; DCM[2][2] = 1.0; } } // DCMNormalise void DCMMotionCompensation(void) { // compensation for rate of change of velocity LR/BF needs GPS velocity but // updates probably too slow? AccV[LR ] += 0.0; AccV[BF] += 0.0; } // DCMMotionCompensation void DCMDriftCorrection(void) { static real32 ScaledOmegaI[3]; //DON'T USE #define USE_DCM_YAW_COMP #ifdef USE_DCM_YAW_COMP static real32 ScaledOmegaP[3]; static real32 YawError[3]; static real32 ErrorCourse; #endif // USE_DCM_YAW_COMP VCross(&RollPitchError[0], &AccV[0], &DCM[2][0]); //adjust the reference ground VScale(&OmegaP[0], &RollPitchError[0], Kp_RollPitch); VScale(&ScaledOmegaI[0], &RollPitchError[0], Ki_RollPitch); VAdd(&OmegaI[0], &OmegaI[0], &ScaledOmegaI[0]); #ifdef USE_DCM_YAW_COMP // Yaw - drift correction based on compass/magnetometer heading HeadingCos = cos(Heading); HeadingSin = sin(Heading); ErrorCourse = ( U[0][0]*HeadingSin ) - ( U[1][0]*HeadingCos ); VScale(YawError, &U[2][0], ErrorCourse ); VScale(&ScaledOmegaP[0], &YawError[0], Kp_Yaw ); VAdd(&OmegaP[0], &OmegaP[0], &ScaledOmegaP[0]); VScale(&ScaledOmegaI[0], &YawError[0], Ki_Yaw ); VAdd(&OmegaI[0], &OmegaI[0], &ScaledOmegaI[0]); #endif // USE_DCM_YAW_COMP } // DCMDriftCorrection void DCMUpdate(void) { static uint8 i, j, k; static real32 op[3]; AccV[BF] = Acc[BF]; AccV[LR] = -Acc[LR]; AccV[UD] = -Acc[UD]; VAdd(&Omega[0], &Gyro[0], &OmegaI[0]); VAdd(&OmegaV[0], &Omega[0], &OmegaP[0]); // MotionCompensation(); U[0][0] = 0.0; U[0][1] = -dT*OmegaV[2]; //-z U[0][2] = dT*OmegaV[1]; // y U[1][0] = dT*OmegaV[2]; // z U[1][1] = 0.0; U[1][2] = -dT*OmegaV[0]; //-x U[2][0] = -dT*OmegaV[1]; //-y U[2][1] = dT*OmegaV[0]; // x U[2][2] = 0.0; for ( i = 0; i < (uint8)3; i++ ) for ( j = 0; j < (uint8)3; j++ ) { for ( k = 0; k < (uint8)3; k++ ) op[k] = DCM[i][k]*U[k][j]; TempM[i][j] = op[0] + op[1] + op[2]; } for ( i = 0; i < (uint8)3; i++ ) for (j = 0; j < (uint8)3; j++ ) DCM[i][j] += TempM[i][j]; } // DCMUpdate void DCMEulerAngles(void) { static uint8 g; for ( g = 0; g < (uint8)3; g++ ) Rate[g] = Gyro[g]; EstAngle[Roll][PremerlaniDCM]= atan2(DCM[2][1], DCM[2][2]); EstAngle[Pitch][PremerlaniDCM] = -asin(DCM[2][0]); EstAngle[Yaw][PremerlaniDCM] = atan2(DCM[1][0], DCM[0][0]); // Est. Rates ??? } // DCMEulerAngles void DoDCM(void) { DCMUpdate(); DCMNormalise(); DCMDriftCorrection(); DCMEulerAngles(); } // DoDCM //___________________________________________________________________________________ // IMU.c // S.O.H. Madgwick // 25th September 2010 // Description: // Quaternion implementation of the 'DCM filter' [Mahony et al.]. // Global variables 'q0', 'q1', 'q2', 'q3' are the quaternion elements representing the estimated // orientation. See my report for an overview of the use of quaternions in this application. // User must call 'IMUupdate()' every sample period and parse calibrated gyroscope ('gx', 'gy', 'gz') // and accelerometer ('ax', 'ay', 'az') data. Gyroscope units are radians/second, accelerometer // units are irrelevant as the vector is normalised. const real32 MKp = 2.0; // proportional gain governs rate of convergence to accelerometer/magnetometer const real32 MKi = 1.0; // integral gain governs rate of convergence of gyroscope biases // 0.005 real32 exInt = 0.0, eyInt = 0.0, ezInt = 0.0; // scaled integral error real32 q0 = 1.0, q1 = 0.0, q2 = 0.0, q3 = 0.0; // quaternion elements representing the estimated orientation void DoMadgwickIMU(real32 gx, real32 gy, real32 gz, real32 ax, real32 ay, real32 az) { static uint8 g; static real32 ReNorm; static real32 vx, vy, vz; static real32 ex, ey, ez; if ( F.AccMagnitudeOK ) { // estimated direction of gravity vx = 2.0*(q1*q3 - q0*q2); vy = 2.0*(q0*q1 + q2*q3); vz = Sqr(q0) - Sqr(q1) - Sqr(q2) + Sqr(q3); // error is sum of cross product between reference direction of field and direction measured by sensor ex = (ay*vz - az*vy); ey = (az*vx - ax*vz); ez = (ax*vy - ay*vx); // integral error scaled integral gain exInt += ex*MKi*dT; eyInt += ey*MKi*dT; ezInt += ez*MKi*dT; // adjusted gyroscope measurements gx += MKp*ex + exInt; gy += MKp*ey + eyInt; gz += MKp*ez + ezInt; // integrate quaternion rate and normalise q0 += (-q1*gx - q2*gy - q3*gz)*dTOn2; q1 += (q0*gx + q2*gz - q3*gy)*dTOn2; q2 += (q0*gy - q1*gz + q3*gx)*dTOn2; q3 += (q0*gz + q1*gy - q2*gx)*dTOn2; // normalise quaternion ReNorm = 1.0 /sqrt(Sqr(q0) + Sqr(q1) + Sqr(q2) + Sqr(q3)); q0 *= ReNorm; q1 *= ReNorm; q2 *= ReNorm; q3 *= ReNorm; MadgwickEulerAngles(MadgwickIMU); } else for ( g = 0; g <(uint8)3; g++) { EstRate[g][MadgwickIMU] = Gyro[g]; EstAngle[g][MadgwickIMU] += EstRate[g][MadgwickIMU]*dT; } } // DoMadgwickIMU //_________________________________________________________________________________ // IMU.c // S.O.H. Madgwick, // 'An Efficient Orientation Filter for Inertial and Inertial/Magnetic Sensor Arrays', // April 30, 2010 #ifdef INC_IMU2 boolean FirstIMU2 = true; real32 BetaIMU2 = 0.033; // const real32 BetaAHRS = 0.041; //Quaternion orientation of earth frame relative to auxiliary frame. real32 AEq_1; real32 AEq_2; real32 AEq_3; real32 AEq_4; //Estimated orientation quaternion elements with initial conditions. real32 SEq_1; real32 SEq_2; real32 SEq_3; real32 SEq_4; void DoMadgwickIMU2(real32 w_x, real32 w_y, real32 w_z, real32 a_x, real32 a_y, real32 a_z) { static uint8 g; //Vector norm. static real32 Renorm; //Quaternion rate from gyroscope elements. static real32 SEqDot_omega_1, SEqDot_omega_2, SEqDot_omega_3, SEqDot_omega_4; //Objective function elements. static real32 f_1, f_2, f_3; //Objective function Jacobian elements. static real32 J_11or24, J_12or23, J_13or22, J_14or21, J_32, J_33; //Objective function gradient elements. static real32 SEqHatDot_1, SEqHatDot_2, SEqHatDot_3, SEqHatDot_4; //Auxiliary variables to avoid reapeated calcualtions. static real32 halfSEq_1, halfSEq_2, halfSEq_3, halfSEq_4; static real32 twoSEq_1, twoSEq_2, twoSEq_3; if ( F.AccMagnitudeOK ) { halfSEq_1 = 0.5*SEq_1; halfSEq_2 = 0.5*SEq_2; halfSEq_3 = 0.5*SEq_3; halfSEq_4 = 0.5*SEq_4; twoSEq_1 = 2.0*SEq_1; twoSEq_2 = 2.0*SEq_2; twoSEq_3 = 2.0*SEq_3; //Compute the quaternion rate measured by gyroscopes. SEqDot_omega_1 = -halfSEq_2*w_x - halfSEq_3*w_y - halfSEq_4*w_z; SEqDot_omega_2 = halfSEq_1*w_x + halfSEq_3*w_z - halfSEq_4*w_y; SEqDot_omega_3 = halfSEq_1*w_y - halfSEq_2*w_z + halfSEq_4*w_x; SEqDot_omega_4 = halfSEq_1*w_z + halfSEq_2*w_y - halfSEq_3*w_x; /* //Normalise the accelerometer measurement. Renorm = 1.0 / sqrt(Sqr(a_x) + Sqr(a_y) + Sqr(a_z)); a_x *= Renorm; a_y *= Renorm; a_z *= Renorm; */ //Compute the objective function and Jacobian. f_1 = twoSEq_2*SEq_4 - twoSEq_1*SEq_3 - a_x; f_2 = twoSEq_1*SEq_2 + twoSEq_3*SEq_4 - a_y; f_3 = 1.0 - twoSEq_2*SEq_2 - twoSEq_3*SEq_3 - a_z; //J_11 negated in matrix multiplication. J_11or24 = twoSEq_3; J_12or23 = 2.0*SEq_4; //J_12 negated in matrix multiplication J_13or22 = twoSEq_1; J_14or21 = twoSEq_2; //Negated in matrix multiplication. J_32 = 2.0*J_14or21; //Negated in matrix multiplication. J_33 = 2.0*J_11or24; //Compute the gradient (matrix multiplication). SEqHatDot_1 = J_14or21*f_2 - J_11or24*f_1; SEqHatDot_2 = J_12or23*f_1 + J_13or22*f_2 - J_32*f_3; SEqHatDot_3 = J_12or23*f_2 - J_33*f_3 - J_13or22*f_1; SEqHatDot_4 = J_14or21*f_1 + J_11or24*f_2; //Normalise the gradient. Renorm = 1.0 / sqrt(Sqr(SEqHatDot_1) + Sqr(SEqHatDot_2) + Sqr(SEqHatDot_3) + Sqr(SEqHatDot_4)); SEqHatDot_1 *= Renorm; SEqHatDot_2 *= Renorm; SEqHatDot_3 *= Renorm; SEqHatDot_4 *= Renorm; //Compute then integrate the estimated quaternion rate. SEq_1 += (SEqDot_omega_1 - (BetaIMU2*SEqHatDot_1))*dT; SEq_2 += (SEqDot_omega_2 - (BetaIMU2*SEqHatDot_2))*dT; SEq_3 += (SEqDot_omega_3 - (BetaIMU2*SEqHatDot_3))*dT; SEq_4 += (SEqDot_omega_4 - (BetaIMU2*SEqHatDot_4))*dT; //Normalise quaternion Renorm = 1.0 / sqrt(Sqr(SEq_1) + Sqr(SEq_2) + Sqr(SEq_3) + Sqr(SEq_4)); SEq_1 *= Renorm; SEq_2 *= Renorm; SEq_3 *= Renorm; SEq_4 *= Renorm; if ( FirstIMU2 ) { //Store orientation of auxiliary frame. AEq_1 = SEq_1; AEq_2 = SEq_2; AEq_3 = SEq_3; AEq_4 = SEq_4; FirstIMU2 = false; } MadgwickEulerAngles(MadgwickIMU2); } else for ( g = 0; g <(uint8)3; g++) { EstRate[g][MadgwickIMU2] = Gyro[g]; EstAngle[g][MadgwickIMU2] += EstRate[g][MadgwickIMU2]*dT; } } // DoMadgwickIMU2 #endif // INC_IMU2 //_________________________________________________________________________________ // AHRS.c // S.O.H. Madgwick // 25th August 2010 // Description: // Quaternion implementation of the 'DCM filter' [Mahoney et al]. Incorporates the magnetic distortion // compensation algorithms from my filter [Madgwick] which eliminates the need for a reference // direction of flux (bx bz) to be predefined and limits the effect of magnetic distortions to yaw // a only. // User must define 'dTOn2' as the (sample period / 2), and the filter gains 'MKp' and 'MKi'. // Global variables 'q0', 'q1', 'q2', 'q3' are the quaternion elements representing the estimated // orientation. See my report for an overview of the use of quaternions in this application. // User must call 'AHRSupdate()' every sample period and parse calibrated gyroscope ('gx', 'gy', 'gz'), // accelerometer ('ax', 'ay', 'az') and magnetometer ('mx', 'my', 'mz') data. Gyroscope units are // radians/second, accelerometer and magnetometer units are irrelevant as the vector is normalised. void DoMadgwickAHRS(real32 gx, real32 gy, real32 gz, real32 ax, real32 ay, real32 az, real32 mx, real32 my, real32 mz) { static uint8 g; static real32 ReNorm; static real32 hx, hy, hz, bx2, bz2, mx2, my2, mz2; static real32 vx, vy, vz, wx, wy, wz; static real32 ex, ey, ez; static real32 q0q0, q0q1, q0q2, q0q3, q1q1, q1q2, q1q3, q2q2, q2q3, q3q3; if ( F.AccMagnitudeOK ) { // auxiliary variables to reduce number of repeated operations q0q0 = q0*q0; q0q1 = q0*q1; q0q2 = q0*q2; q0q3 = q0*q3; q1q1 = q1*q1; q1q2 = q1*q2; q1q3 = q1*q3; q2q2 = q2*q2; q2q3 = q2*q3; q3q3 = q3*q3; ReNorm = 1.0 / sqrt( Sqr( mx ) + Sqr( my ) + Sqr( mz ) ); mx *= ReNorm; my *= ReNorm; mz *= ReNorm; mx2 = 2.0*mx; my2 = 2.0*my; mz2 = 2.0*mz; // compute reference direction of flux hx = mx2*(0.5 - q2q2 - q3q3) + my2*(q1q2 - q0q3) + mz2*(q1q3 + q0q2); hy = mx2*(q1q2 + q0q3) + my2*( 0.5 - q1q1 - q3q3) + mz2*(q2q3 - q0q1); hz = mx2*(q1q3 - q0q2) + my2*(q2q3 + q0q1) + mz2*( 0.5 - q1q1 - q2q2 ); bx2 = 2.0*sqrt( Sqr( hx ) + Sqr( hy ) ); bz2 = 2.0*hz; // estimated direction of gravity and flux (v and w) vx = 2.0*(q1q3 - q0q2); vy = 2.0*(q0q1 + q2q3); vz = q0q0 - q1q1 - q2q2 + q3q3; wx = bx2*(0.5 - q2q2 - q3q3) + bz2*(q1q3 - q0q2); wy = bx2*(q1q2 - q0q3) + bz2*( q0q1 + q2q3 ); wz = bx2*(q0q2 + q1q3) + bz2*( 0.5 - q1q1 - q2q2 ); // error is sum of cross product between reference direction of fields and direction measured by sensors ex = (ay*vz - az*vy) + (my*wz - mz*wy); ey = (az*vx - ax*vz) + (mz*wx - mx*wz); ez = (ax*vy - ay*vx) + (mx*wy - my*wx); // integral error scaled integral gain exInt += ex*MKi*dT; eyInt += ey*MKi*dT; ezInt += ez*MKi*dT; // adjusted gyroscope measurements gx += MKp*ex + exInt; gy += MKp*ey + eyInt; gz += MKp*ez + ezInt; // integrate quaternion rate and normalise q0 += (-q1*gx - q2*gy - q3*gz)*dTOn2; q1 += (q0*gx + q2*gz - q3*gy)*dTOn2; q2 += (q0*gy - q1*gz + q3*gx)*dTOn2; q3 += (q0*gz + q1*gy - q2*gx)*dTOn2; // normalise quaternion ReNorm = 1.0 / sqrt(Sqr(q0) + Sqr(q1) + Sqr(q2) + Sqr(q3)); q0 *= ReNorm; q1 *= ReNorm; q2 *= ReNorm; q3 *= ReNorm; MadgwickEulerAngles(MadgwickAHRS); } else for ( g = 0; g <(uint8)3; g++) { EstRate[g][MadgwickAHRS] = Gyro[g]; EstAngle[g][MadgwickAHRS] += EstRate[g][MadgwickAHRS]*dT; } } // DoMadgwickAHRS void MadgwickEulerAngles(uint8 S) { EstAngle[Roll][S] = atan2(2.0*q2*q3 - 2.0*q0*q1, 2.0*Sqr(q0) + 2.0*Sqr(q3) - 1.0); EstAngle[Pitch][S] = asin(2.0*q1*q2 - 2.0*q0*q2); EstAngle[Yaw][S] = atan2(2.0*q1*q2 - 2.0*q0*q3, 2.0*Sqr(q0) + 2.0*Sqr(q1) - 1.0); } // MadgwickEulerAngles //_________________________________________________________________________________ // Kalman Filter originally authored by Tom Pycke // http://tom.pycke.be/mav/71/kalman-filtering-of-imu-data real32 AngleKF[2] = {0,0}; real32 BiasKF[2] = {0,0}; real32 P00[2] = {0,0}; real32 P01[2] = {0,0}; real32 P10[2] = {0,0}; real32 P11[2] = {0,0}; real32 KalmanFilter(uint8 a, real32 NewAngle, real32 NewRate) { // Q is a 2x2 matrix that represents the process covariance noise. // In this case, it indicates how much we trust the accelerometer // relative to the gyros. const real32 AngleQ = 0.003; const real32 GyroQ = 0.009; // R represents the measurement covariance noise. In this case, // it is a 1x1 matrix that says that we expect AngleR rad jitter // from the accelerometer. const real32 AngleR = GYRO_PROP_NOISE; static real32 y, S; static real32 K0, K1; AngleKF[a] += (NewRate - BiasKF[a])*dT; P00[a] -= (( P10[a] + P01[a] ) + AngleQ )*dT; P01[a] -= P11[a]*dT; P10[a] -= P11[a]*dT; P11[a] += GyroQ*dT; y = NewAngle - AngleKF[a]; S = 1.0 / ( P00[a] + AngleR ); K0 = P00[a]*S; K1 = P10[a]*S; AngleKF[a] += K0*y; BiasKF[a] += K1*y; P00[a] -= K0*P00[a]; P01[a] -= K0*P01[a]; P10[a] -= K1*P00[a]; P11[a] -= K1*P01[a]; return ( AngleKF[a] ); } // KalmanFilter void DoKalman(void) { // NO YAW ANGLE ESTIMATE EstAngle[Roll][Kalman] = KalmanFilter(Roll, asin(-Acc[LR]), Gyro[Roll]); EstAngle[Pitch][Kalman] = KalmanFilter(Pitch, asin(Acc[BF]), Gyro[Pitch]); EstRate[Roll][Kalman] = Gyro[Roll]; EstRate[Pitch][Kalman] = Gyro[Pitch]; } // DoKalman //_________________________________________________________________________________ // MultWii // Original code by Alex at: http://radio-commande.com/international/triwiicopter-design/ // simplified IMU based on Kalman Filter // inspired from http://starlino.com/imu_guide.html // and http://www.starlino.com/imu_kalman_arduino.html // with this algorithm, we can get absolute angles for a stable mode integration real32 AccMW[3] = {0.0, 0.0, 1.0}; // init acc in stable mode real32 GyroMW[3] = {0.0, 0.0, 0.0}; // R obtained from last estimated value and gyro movement real32 Axz, Ayz; // angles between projection of R on XZ/YZ plane and Z axis void DoMultiWii(void) { // V1.6 NO YAW ANGLE ESTIMATE const real32 GyroWt = 50.0; // gyro weight/smoothing factor const real32 GyroWtR = 1.0 / GyroWt; if ( Acc[UD] < 0.0 ) { // check not inverted if ( F.AccMagnitudeOK ) { Ayz = atan2( AccMW[LR], AccMW[UD] ) + Gyro[Roll]*dT; Axz = atan2( AccMW[BF], AccMW[UD] ) + Gyro[Pitch]*dT; } else { Ayz += Gyro[Roll]*dT; Axz += Gyro[Pitch]*dT; } // reverse calculation of GyroMW from Awz angles, // for formulae deduction see http://starlino.com/imu_guide.html GyroMW[Roll] = sin( Ayz ) / sqrt( 1.0 + Sqr( cos( Ayz ) )*Sqr( tan( Axz ) ) ); GyroMW[Pitch] = sin( Axz ) / sqrt( 1.0 + Sqr( cos( Axz ) )*Sqr( tan( Ayz ) ) ); GyroMW[Yaw] = sqrt( fabs( 1.0 - Sqr( GyroMW[Roll] ) - Sqr( GyroMW[Pitch] ) ) ); //combine accelerometer and gyro readings AccMW[LR] = ( -Acc[LR] + GyroWt*GyroMW[Roll] )*GyroWtR; AccMW[BF] = ( Acc[BF] + GyroWt*GyroMW[Pitch] )*GyroWtR; AccMW[UD] = ( -Acc[UD] + GyroWt*GyroMW[Yaw] )*GyroWtR; } EstAngle[Roll][MultiWii] = Ayz; EstAngle[Pitch][MultiWii] = Axz; // EstRate[Roll][MultiWii] = GyroMW[Roll]; // EstRate[Pitch][MultiWii] = GyroMW[Pitch]; EstRate[Roll][MultiWii] = Gyro[Roll]; EstRate[Pitch][MultiWii] = Gyro[Pitch]; } // DoMultiWii //_________________________________________________________________________________ void AttitudeTest(void) { TxString("\r\nAttitude Test\r\n"); GetAttitude(); TxString("\r\ndT \t"); TxVal32(dT*1000.0, 3, 0); TxString(" Sec.\r\n\r\n"); if ( GyroType == IRSensors ) { TxString("IR Sensors:\r\n"); TxString("\tRoll \t"); TxVal32(IR[Roll]*100.0, 2, HT); TxNextLine(); TxString("\tPitch\t"); TxVal32(IR[Pitch]*100.0, 2, HT); TxNextLine(); TxString("\tZ \t"); TxVal32(IR[Yaw]*100.0, 2, HT); TxNextLine(); TxString("\tMin/Max\t"); TxVal32(IRMin*100.0, 2, HT); TxVal32(IRMax*100.0, 2, HT); TxString("\tSwing\t"); TxVal32(IRSwing*100.0, 2, HT); TxNextLine(); } else { TxString("Gyro, Compensated, Max Delta(Deg./Sec.):\r\n"); TxString("\tRoll \t"); TxVal32(Gyro[Roll]*MILLIANGLE, 3, HT); TxVal32(Rate[Roll]*MILLIANGLE, 3, HT); TxVal32(GyroNoise[Roll]*MILLIANGLE,3, 0); TxNextLine(); TxString("\tPitch\t"); TxVal32(Gyro[Pitch]*MILLIANGLE, 3, HT); TxVal32(Rate[Pitch]*MILLIANGLE, 3, HT); TxVal32(GyroNoise[Pitch]*MILLIANGLE,3, 0); TxNextLine(); TxString("\tYaw \t"); TxVal32(Gyro[Yaw]*MILLIANGLE, 3, HT); TxVal32(Rate[Yaw]*MILLIANGLE, 3, HT); TxVal32(GyroNoise[Yaw]*MILLIANGLE, 3, 0); TxNextLine(); TxString("Accelerations , peak change(G):\r\n"); TxString("\tB->F \t"); TxVal32(Acc[BF]*1000.0, 3, HT); TxVal32( AccNoise[BF]*1000.0, 3, 0); TxNextLine(); TxString("\tL->R \t"); TxVal32(Acc[LR]*1000.0, 3, HT); TxVal32( AccNoise[LR]*1000.0, 3, 0); TxNextLine(); TxString("\tU->D \t"); TxVal32(Acc[UD]*1000.0, 3, HT); TxVal32( AccNoise[UD]*1000.0, 3, 0); TxNextLine(); } if ( CompassType != HMC6352 ) { TxString("Magnetic:\r\n"); TxString("\tX \t"); TxVal32(Mag[Roll].V, 0, 0); TxNextLine(); TxString("\tY \t"); TxVal32(Mag[Pitch].V, 0, 0); TxNextLine(); TxString("\tZ \t"); TxVal32(Mag[Yaw].V, 0, 0); TxNextLine(); } TxString("Heading: \t"); TxVal32(Make2Pi(Heading)*MILLIANGLE, 3, 0); TxNextLine(); } // AttitudeTest void InitAttitude(void) { static uint8 a, s; #ifdef INC_IMU2 FirstIMU2 = true; BetaIMU2 = sqrt(0.75) * GyroNoiseRadian[GyroType]; //Quaternion orientation of earth frame relative to auxiliary frame. AEq_1 = 1.0; AEq_2 = 0.0; AEq_3 = 0.0; AEq_4 = 0.0; //Estimated orientation quaternion elements with initial conditions. SEq_1 = 1.0; SEq_2 = 0.0; SEq_3 = 0.0; SEq_4 = 0.0; #endif // INC_IMU2 for ( a = 0; a < (uint8)2; a++ ) AngleCF[a] = AngleKF[a] = BiasKF[a] = F0[a] = F1[a] = F2[a] = 0.0; for ( a = 0; a < (uint8)3; a++ ) for ( s = 0; s < MaxAttitudeScheme; s++ ) EstAngle[a][s] = EstRate[a][s] = 0.0; } // InitAttitude