forked from terejanu/AdaptiveGaussianSumFilter
-
Notifications
You must be signed in to change notification settings - Fork 0
/
runFilters.m
executable file
·261 lines (221 loc) · 9.35 KB
/
runFilters.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
function [xt,F1,F2,F4,PF] = runFilters(iRun,model,prior,time,measAvail,pf,inc,myfilter)
%
% Run all the 4 filters for one particular MC run. Filters:
% 1. no update when measurements are available
% 2. Bayesian classical update
% 4. Continuous update based on FPKE
%
% for each Fx/PF there will be saved:
% Fx.x{k} - estimate first moment (mean) - over time
% Fx.P{k} - estimate second moment (covariance) - over time
% Fx.w{k} - weights for gaussian components - over time
% Fx.ll{k} - log-likelihood of the particles - over time
% Fx.pdf{k}.dim(j).mmX - pdf for dim j (1:model.fn) - X axis
% Fx.pdf{k}.dim(j).mmY - pdf for dim j (1:model.fn) - Y axis
% Fx.pdf{k}.dim(j).mmZ - pdf for dim j (1:model.fn) - Z axis
%
% Gabriel Terejanu ([email protected])
%% ------------------------------------------------------------------------
% init
%--------------------------------------------------------------------------
path(path, genpath(inc));
opt = odeset('reltol',1e-8,'abstol',1e-8);
fprintf('- Run # %d\n',iRun);
%% ------------------------------------------------------------------------
% create truth & measurement
%--------------------------------------------------------------------------
fprintf(' - create truth & measurements\n');
xt = cell(1, time.nSteps);
y_meas = cell(1, time.nSteps);
% select a gaussian component
rand('state',iRun*100);
gs_sel = 1;
u = rand;
u_total = 0;
for j = 1 : prior.n
if ((u >= u_total) && (u < u_total + prior.weig(j)))
gs_sel = j;
break;
end
u_total = u_total + prior.weig(j);
end
% draw a sample from the chosen gaussian component
randn('state',iRun*100);
xt{1} = prior.mu{gs_sel} + chol(prior.sig{gs_sel})' * randn(model.fn,1);
y_meas{1} = feval(model.hx,xt{1}) + chol(model.R)' * randn(model.hn,1);
% get the trajectory of the sample over time
for k = 2 : time.nSteps
Ttmp = [time.tspan(k-1) time.tspan(k)]; % integration period
xt{k} = sde_int(model, time, Ttmp, xt{k-1});
y_meas{k} = feval(model.hx,xt{k}) + chol(model.R)' * randn(model.hn,1);
end
%% ------------------------------------------------------------------------
% PF - for particle filters
%--------------------------------------------------------------------------
fprintf(' - create initial particles\n');
% generate i.c. samples
u_pf = rand(1,pf.no_particles);
tu_pf = 0;
X_pf = [];
for i = 1 : prior.n
% for each gaussian component draw samples dictated by its weight
% magnitude
ind = find((u_pf >= tu_pf) & (u_pf < tu_pf + prior.weig(i)));
Xtmp_pf = repmat(prior.mu{i},1,length(ind)) + chol(prior.sig{i})' * randn(model.fn,length(ind));
% collect all the samples and advance
X_pf = [X_pf Xtmp_pf];
tu_pf = tu_pf + prior.weig(i);
end
w_pf = ones(1, pf.no_particles) / pf.no_particles;
%% ------------------------------------------------------------------------
% compute initial estimates
%--------------------------------------------------------------------------
F1.w{1} = prior.weig;
F2.w{1} = prior.weig;
F4.w{1} = prior.weig;
[F1.x{1},F1.P{1},F1.ll{1}] = getGSdata(F1.w{1},prior.mu,prior.sig,X_pf,w_pf);
[F2.x{1},F2.P{1},F2.ll{1}] = getGSdata(F2.w{1},prior.mu,prior.sig,X_pf,w_pf);
[F4.x{1},F4.P{1},F4.ll{1}] = getGSdata(F4.w{1},prior.mu,prior.sig,X_pf,w_pf);
[PF.x{1},PF.P{1}] = getPFdata(X_pf,w_pf);
save_weights = prior.weig;
%% ------------------------------------------------------------------------
% init filters
%--------------------------------------------------------------------------
mu = prior.mu;
sig = prior.sig;
save_mus{1}.mus = mu;
%% ------------------------------------------------------------------------
% RUN FILTERS
%--------------------------------------------------------------------------
for k = 2 : time.nSteps
fprintf(' - run filters tStep %d / %d\n', k, time.nSteps);
% save data
mu_old = mu;
sig_old = sig;
% get the integration interval
Ttmp = [time.tspan(k-1) time.tspan(k)];
%% ------------------------------------------------------------------------
% PF - propagate particles
%--------------------------------------------------------------------------
fprintf(' - Particle Filter\n');
% PF - time update
X_pf = sde_int(model, time, Ttmp, X_pf);
% PF - measurement update
if (measAvail(k) == 1)
[X_pf, w_pf] = PF_meas_update(X_pf, w_pf, model, pf, y_meas{k});
end;
% PF - compute estimates
[PF.x{k},PF.P{k}] = getPFdata(X_pf, w_pf);
%% ------------------------------------------------------------------------
% propagate and update each gaussian component using Kalman Filter
%--------------------------------------------------------------------------
fprintf(' - Kalman Filter\n');
for j = 1 : prior.n
% time update
[mu{j}, sig{j}] = feval([myfilter.name '_time_update'], myfilter, model, mu{j}, sig{j}, Ttmp, opt);
% do update
if (measAvail(k) == 1)
[mu_up{j}, sig_up{j}, z_mu{j}, z_sig{j}] = feval([myfilter.name '_measurement_update'], myfilter, model, mu{j}, sig{j}, Ttmp, opt, y_meas{k});
else
mu_up{j} = mu{j};
sig_up{j} = sig{j};
end
end
%% ------------------------------------------------------------------------
% GS1 - no weight update
%--------------------------------------------------------------------------
fprintf(' - Weights GS1\n');
F1.w{k} = F1.w{k-1};
%% ------------------------------------------------------------------------
% GS2 - Alspach - classic weight update
%--------------------------------------------------------------------------
fprintf(' - Weights GS2\n');
if (measAvail(k) == 1)
for j = 1 : prior.n
F2.w{k}(j) = F2.w{k-1}(j)*getLikelihood(y_meas{k} - z_mu{j}, z_sig{j} + model.R);
end
if (sum(F2.w{k}) > 0)
F2.w{k} = F2.w{k} ./ sum(F2.w{k});
else
F2.w{k} = F2.w{k-1};
end;
else
F2.w{k} = F2.w{k-1};
end
%% ------------------------------------------------------------------------
% GS4 - Continuous Update - using FPKE
%--------------------------------------------------------------------------
fprintf(' - Weights GS4\n');
switch myfilter.integration_method
case 'IUT'
% compute matrix H - Unscented Transformation
[M,N] = ComputeFPE_QP_UT_noPG(F4.w{k-1}, mu, sig, model, time, k);
case 'GQ_perComp'
% compute matrix H - using Gaussian Quadrature with quadrature
% points for each gaussian component
% LL = ComputeFPE_QP_GQ_perComp(F4.w{k-1}, mu, sig, model, time, k, myfilter);
% H = LL+eye(size(LL));
case 'GQ_all'
% compute matrix H - using Gaussian Quadrature with
% quadrature points that cover the entire domain
% LL = ComputeFPE_QP_GQ_all(F4.w{k-1}, mu, sig, model, time, k, myfilter);
% H = LL+eye(size(LL));
otherwise
error('Integration method unknown');
end;
% get the weights
Aeq = ones(1,prior.n); beq =1; Ain = -eye(prior.n); bin = zeros(prior.n,1);
xx = sdpvar(prior.n,1);
xx_old = reshape(F4.w{k-1},prior.n,1);
cons = set(Aeq*xx == beq) + set(Ain*xx < bin);
assign(xx, xx_old);
optyalmip = sdpsettings('solver','sedumi','verbose',0,'usex0',1);
diagnostic = solvesdp(cons,1/2*xx'*M*xx + xx'*N*xx_old, optyalmip);
F4.w{k} = double(xx);
saveF4w = F4.w{k};
% do the classic weight update GS2
if (measAvail(k) == 1)
for j = 1 : prior.n
F4.w{k}(j) = F4.w{k}(j)*getLikelihood(y_meas{k} - z_mu{j}, z_sig{j} + model.R);
end
if (sum(F4.w{k}) > 0)
F4.w{k} = F4.w{k} ./ sum(F4.w{k});
else
F4.w{k} = saveF4w;
end;
end
save_weights = [save_weights; F4.w{k}'];
F2.w{k}
F4.w{k}'
%% ------------------------------------------------------------------------
% Store estimates
%--------------------------------------------------------------------------
mu = mu_up;
sig = sig_up;
save_mus{k}.mus = mu;
%% ------------------------------------------------------------------------
% Compute estimates
%--------------------------------------------------------------------------
[F1.x{k},F1.P{k},F1.ll{k}] = getGSdata(F1.w{k},mu,sig,X_pf,w_pf);
[F2.x{k},F2.P{k},F2.ll{k}] = getGSdata(F2.w{k},mu,sig,X_pf,w_pf);
[F4.x{k},F4.P{k},F4.ll{k}] = getGSdata(F4.w{k},mu,sig,X_pf,w_pf);
%% ------------------------------------------------------------------------
% Plot
%--------------------------------------------------------------------------
plot3(X_pf(1,:),X_pf(2,:),X_pf(3,:),'.r'); hold on;
for j = 1 : prior.n
myX = zeros(model.fn,k);
for i = 1 : k
myX(:,i) = save_mus{i}.mus{j};
end;
error_ellipse(sig{j},mu{j});
alpha(F4.w{k}(j));
plot3(myX(1,:),myX(2,:),myX(3,:),'LineWidth',3);
end;
hold off;grid on;view(355,30);
xlabel('x');ylabel('y');zlabel('z');
drawnow;
saveas(gcf,['images/' num2str(k,'%03d')], 'bmp');
saveas(gcf,['images/' num2str(k,'%03d')], 'fig');
pause(0.2);
end