mirror of https://github.com/ArduPilot/ardupilot
203 lines
7.3 KiB
Matlab
203 lines
7.3 KiB
Matlab
%% Set initial conditions
|
|
clear all;
|
|
dtSlow = 1/50;
|
|
dtFast = 1/1000;
|
|
rateMult = round(dtSlow/dtFast);
|
|
duration = 60;
|
|
indexLimitSlow = round(duration/dtSlow);
|
|
indexLimitFast = indexLimitSlow*rateMult;
|
|
|
|
% create data logging variables
|
|
gimbal.time = zeros(1,indexLimitFast);
|
|
gimbal.euler = zeros(3,indexLimitFast);
|
|
gimbal.eulerTruth = zeros(3,indexLimitFast);
|
|
gimbal.eulerError = zeros(3,indexLimitFast);
|
|
|
|
% Use a random initial truth orientation
|
|
phiInit = 0.1*randn;
|
|
thetaInit = 0.1*randn;
|
|
psiInit = 2*pi*rand - pi;
|
|
quatTruth = EulToQuat([phiInit,thetaInit,psiInit]);% [1;0.05*randn;0.05*randn;2*(rand-0.5)];
|
|
quatLength = sqrt(quatTruth(1)^2 + quatTruth(2)^2 + quatTruth(3)^2 + quatTruth(4)^2);
|
|
quatTruth = quatTruth / quatLength;
|
|
TsnTruth = Quat2Tbn(quatTruth);
|
|
|
|
% define the earths truth magnetic field
|
|
declTruth = 10*pi/180;
|
|
magEarthTruth = [0.25*cos(declTruth);0.25*sin(declTruth);-0.5];
|
|
|
|
% define the declination parameter assuming 2deg RMS error - this would be
|
|
% obtained from the main EKF to take advantage of in-flight learning
|
|
declParam = declTruth + 2*pi/180*randn;
|
|
|
|
% define the magnetometer bias errors
|
|
magMeasBias = 0.02*[randn;randn;randn];
|
|
|
|
% Define IMU bias errors and noise
|
|
gyroBias = 1*pi/180*[randn;randn;randn];
|
|
accBias = 0.05*[randn;randn;randn];
|
|
gyroNoise = 0.01;
|
|
accNoise = 0.05;
|
|
|
|
% define the state covariances with the exception of the quaternion covariances
|
|
Sigma_velNED = 0.5; % 1 sigma uncertainty in horizontal velocity components
|
|
Sigma_dAngBias = 1*pi/180*dtSlow; % 1 Sigma uncertainty in delta angle bias
|
|
Sigma_angErr = 1; % 1 Sigma uncertainty in angular misalignment (rad)
|
|
covariance = single(diag([Sigma_angErr*[1;1;1];Sigma_velNED*[1;1;1];Sigma_dAngBias*[1;1;1]]).^2);
|
|
|
|
% Initialise truth trajectory variables
|
|
% fly a CCW circle with constant gimbal angles
|
|
gPsiInit = 20*pi/180; % gimbal yaw
|
|
gThetaInit = 0; % gimbal pitch
|
|
gPhiInit = 0; % gimbal roll
|
|
psiTruth = psiInit;
|
|
radius = 20;
|
|
gndSpd = 5;
|
|
trackAngTruth = -pi;
|
|
centripAccelMag = gndSpd/radius*gndSpd;
|
|
gravAccel = [0;0;-9.81];
|
|
|
|
%% Main Loop
|
|
hdgAlignedEKF=0;
|
|
hdgAlignedGimbal=0;
|
|
slowIndex = 0;
|
|
delAngFast = [0;0;0];
|
|
delVelFast = [0;0;0];
|
|
delAngSlow = [0;0;0];
|
|
delVelSlow = [0;0;0];
|
|
prevAngRateMeas = [0;0;0];
|
|
prevAccelMeas = [0;0;0];
|
|
quatFast = [1;0;0;0];
|
|
quatFastSaved = quatFast;
|
|
angRateBiasEKF = [0;0;0];
|
|
quatEKF = [1;0;0;0];
|
|
for fastIndex = 1:indexLimitFast % 1000 Hz gimbal prediction loop
|
|
time = dtFast*fastIndex;
|
|
% Calculate Truth Data
|
|
% Need to replace this with a full kinematic model or test data
|
|
% calculate truth angular rates - we don't start maneouvring until
|
|
% heading alignment is complete
|
|
psiRateTruth = gndSpd/radius*hdgAlignedEKF;
|
|
angRateTruth = [0;0;psiRateTruth]; % constant yaw rate
|
|
|
|
% calculate yaw and track angles
|
|
psiTruth = psiTruth + psiRateTruth*dtFast;
|
|
trackAngTruth = trackAngTruth + psiRateTruth*dtFast;
|
|
|
|
% Cacluate truth quternion
|
|
quatTruth = EulToQuat([phiInit,thetaInit,psiTruth]);
|
|
|
|
% Calculate truth rotaton from sensor to NED
|
|
TsnTruth = Quat2Tbn(quatTruth);
|
|
|
|
% calculate truth accel vector
|
|
centripAccel = centripAccelMag*[-sin(trackAngTruth);cos(trackAngTruth);0];
|
|
accelTruth = transpose(TsnTruth)*(gravAccel + centripAccel);
|
|
|
|
% calculate truth velocity vector
|
|
truthVel = gndSpd*[cos(trackAngTruth);sin(trackAngTruth);0];
|
|
|
|
% synthesise sensor measurements
|
|
% Synthesise IMU measurements, adding bias and noise
|
|
angRateMeas = angRateTruth + gyroBias + gyroNoise*[randn;randn;randn];
|
|
accelMeas = accelTruth + accBias + accNoise*[randn;randn;randn];
|
|
|
|
% synthesise velocity measurements
|
|
measVel = truthVel;
|
|
|
|
% synthesise gimbal angles
|
|
gPhi = 0;
|
|
gTheta = 0;
|
|
gPsi = gPsiInit;
|
|
|
|
% Define rotation from magnetometer to sensor using a 312 rotation sequence
|
|
TmsTruth = calcTms(gPhi,gPsi,gTheta);
|
|
|
|
% calculate rotation from NED to magnetometer axes Tnm = Tsm * Tns
|
|
TnmTruth = transpose(TmsTruth) * transpose(TsnTruth);
|
|
|
|
% synthesise magnetometer measurements adding sensor bias
|
|
magMeas = TnmTruth*magEarthTruth + magMeasBias;
|
|
|
|
% integrate the IMU measurements to produce delta angles and velocities
|
|
% using a trapezoidal integrator
|
|
if isempty(prevAngRateMeas)
|
|
prevAngRateMeas = angRateMeas;
|
|
end
|
|
if isempty(prevAccelMeas)
|
|
prevAccelMeas = accelMeas;
|
|
end
|
|
delAngFast = delAngFast + 0.5*(angRateMeas + prevAngRateMeas)*dtFast;
|
|
delVelFast = delVelFast + 0.5*(accelMeas + prevAccelMeas)*dtFast;
|
|
prevAngRateMeas = angRateMeas;
|
|
prevAccelMeas = accelMeas;
|
|
|
|
% Run an attitude prediction calculation at 1000Hz
|
|
% Convert the rotation vector to its equivalent quaternion
|
|
% using a first order approximation after applying the correction for
|
|
% gyro bias using bias estimates from the EKF
|
|
deltaQuat = [1;0.5*(angRateMeas - angRateBiasEKF)*dtFast];
|
|
% Update the quaternions by rotating from the previous attitude through
|
|
% the delta angle rotation quaternion
|
|
quatFast = QuatMult(quatFast,deltaQuat);
|
|
% Normalise the quaternions
|
|
quatFast = NormQuat(quatFast);
|
|
% log the high rate data
|
|
eulLogFast(:,fastIndex) = QuatToEul(quatFast);
|
|
|
|
% every 20msec we send them to the EKF computer and reset
|
|
% the total
|
|
% we also save a copy of the quaternion from our high rate prediction
|
|
if (rem(fastIndex,rateMult) == 0)
|
|
delAngSlow = delAngFast;
|
|
delVelSlow = delVelFast;
|
|
delAngFast = [0;0;0];
|
|
delVelFast = [0;0;0];
|
|
quatFastSaved = quatFast;
|
|
end
|
|
|
|
% run the 50Hz EKF loop but do so 5 msec behind the
|
|
% data transmission to simulate the effect of transmission and
|
|
% computational delays
|
|
if (rem(fastIndex,rateMult) == 5)
|
|
slowIndex = slowIndex + 1;
|
|
[quatEKF,angRateBiasEKF,EKFlogs,hdgAlignedEKF] = calcEKF(delAngSlow,delVelSlow,measVel,gPhi,gPsi,gTheta,magMeas,declParam,time,dtSlow,slowIndex,indexLimitSlow);
|
|
end
|
|
|
|
% Correct Gimbal attitude usng EKF data
|
|
% Assume the gimbal controller receive the EKF solution 10 msec after
|
|
% it sent the sensor data
|
|
if (rem(fastIndex,rateMult) == 10)
|
|
% calculate the quaternion from the EKF corrected attitude to the
|
|
% attitude calculated using the local fast prediction algorithm
|
|
deltaQuatFast = QuatDivide(quatEKF,quatFastSaved);
|
|
% apply this correction to the fast solution at the current time
|
|
% step (this can be applied across several steps to smooth the
|
|
% output if required)
|
|
quatFast = QuatMult(quatFast,deltaQuatFast);
|
|
% normalise the resultant quaternion
|
|
quatFast = NormQuat(quatFast);
|
|
% flag when the gimbals own heading is aligned
|
|
hdgAlignedGimbal = hdgAlignedEKF;
|
|
end
|
|
|
|
% Log gimbal data
|
|
gimbal.time(fastIndex) = time;
|
|
gimbal.euler(:,fastIndex) = QuatToEul(quatFast);
|
|
gimbal.eulerTruth(:,fastIndex) = QuatToEul(quatTruth);
|
|
if (hdgAlignedGimbal)
|
|
gimbal.eulerError(:,fastIndex) = gimbal.euler(:,fastIndex) - gimbal.eulerTruth(:,fastIndex);
|
|
if (gimbal.eulerError(3,fastIndex) > pi)
|
|
gimbal.eulerError(3,fastIndex) = gimbal.eulerError(3,fastIndex) - 2*pi;
|
|
elseif (gimbal.eulerError(3,fastIndex) < -pi)
|
|
gimbal.eulerError(3,fastIndex) = gimbal.eulerError(3,fastIndex) + 2*pi;
|
|
end
|
|
else
|
|
gimbal.eulerError(:,fastIndex) = [NaN;NaN;NaN];
|
|
end
|
|
|
|
end
|
|
|
|
%% Generate Plots
|
|
close all;
|
|
PlotData; |