demo_GPR01.m 10.1 KB
Newer Older
1
function demo_GPR01
2 3
% Gaussian processs regression (GPR) used for task adaptation, with DS-GMR used to 
% retrieve continuous movements.
Sylvain Calinon's avatar
Sylvain Calinon committed
4
% 
5 6 7 8
% Writing code takes time. Polishing it and making it available to others takes longer! 
% If some parts of the code were useful for your research of for a better understanding 
% of the algorithms, please reward the authors by citing the related publications, 
% and consider making your own research available in this way.
Sylvain Calinon's avatar
Sylvain Calinon committed
9 10 11
%
% @article{Calinon15,
%   author="Calinon, S.",
12 13 14
%   title="A Tutorial on Task-Parameterized Movement Learning and Retrieval",
%   journal="Intelligent Service Robotics",
%   year="2015"
Sylvain Calinon's avatar
Sylvain Calinon committed
15
% }
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
% 
% Copyright (c) 2015 Idiap Research Institute, http://idiap.ch/
% Written by Sylvain Calinon, http://calinon.ch/
% 
% This file is part of PbDlib, http://www.idiap.ch/software/pbdlib/
% 
% PbDlib is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License version 3 as
% published by the Free Software Foundation.
% 
% PbDlib 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 PbDlib. If not, see <http://www.gnu.org/licenses/>.
33 34 35

addpath('./m_fcts/');

Sylvain Calinon's avatar
Sylvain Calinon committed
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
%% Parameters
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
model.nbStates = 3; %Number of Gaussians in the GMM
model.nbFrames = 2; %Number of candidate frames of reference
model.nbVar = 3; %Dimension of the datapoints in the dataset (here: t,x1,x2)
model.dt = 0.01; %Time step
model.kP = 100; %Stiffness gain (for DS-GMR)
model.kV = (2*model.kP)^.5; %Damping gain (with ideal underdamped damping ratio)
nbRepros = 8; %Number of reproductions with new situations randomly generated


%% Load 3rd order tensor data
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
disp('Load motion data...');
% The MAT file contains a structure 's' with the multiple demonstrations. 's(n).Data' is a matrix data for
% sample n (with 's(n).nbData' datapoints). 's(n).p(m).b' and 's(n).p(m).A' contain the position and
% orientation of the m-th candidate coordinate system for this demonstration. 'Data' contains the observations
% in the different frames. It is a 3rd order tensor of dimension D x P x N, with D=3 the dimension of a
% datapoint, P=2 the number of candidate frames, and N=200x4 the number of datapoints in a trajectory (200)
% multiplied by the number of demonstrations (nbSamples=5).
load('data/DataLQR01.mat');


%% Transformation of 'Data' to learn the path of the spring-damper system
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
nbD = s(1).nbData;
nbVarOut = model.nbVar - 1;
%Create transformation matrix to compute [X; DX; DDX]
D = (diag(ones(1,nbD-1),-1)-eye(nbD)) / model.dt;
D(end,end) = 0;
%Create transformation matrix to compute XHAT = X + DX*kV/kP + DDX/kP
K1d = [1, model.kV/model.kP, 1/model.kP];
K = kron(K1d,eye(nbVarOut));
%Create data with XHAT instead of X, see Eq. (4.0.2) in doc/TechnicalReport.pdf
Data = s(1).Data0(1,:);
for n=1:nbSamples
	DataTmp = s(n).Data0(2:end,:);
	s(n).Data = K * [DataTmp; DataTmp*D; DataTmp*D*D];
	Data = [Data; s(n).Data]; %Data is a matrix of size M*D x T (stacking the different trajectory samples)
end


%% Prepare input-output data for GPR from GMM encoding
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
fprintf('Parameters estimation of GPR with GMM encoding:');
%GMM encoding for each trajectory sample, learned at once by stacking the different trajectory samples in 'Data' matrix of size M*D x T
model.nbVar = size(Data,1); %Temporary modification of nbVar (for stacked data training)
model = init_GMM_timeBased(Data, model);
model = EM_GMM(Data, model);
model.nbVar = size(s(1).p(1).A,1); %Setting back the initial nbVar
for n=1:nbSamples
	id = (n-1)*2+2:n*2+1;
	s(n).Priors = model.Priors;
	s(n).Mu = model.Mu([1,id],:);
	s(n).Sigma = model.Sigma([1,id],[1,id],:);
	%Regularization of Sigma
	for i=1:model.nbStates
		[V,D] = eig(s(n).Sigma(:,:,i));
		s(n).Sigma(:,:,i) = V * max(D,1E-3) * V';
	end
	%Set query point vector (position and orientation of the two objects)
	s(n).DataIn = [s(n).p(1).b(2:3); s(n).p(1).A(2:3,3); s(n).p(2).b(2:3); s(n).p(2).A(2:3,3)];
	model.DataIn(:,n) = s(n).DataIn;
	%Set model output vector (Mu and Sigma)
	model.DataOut(:,n) = [reshape(s(n).Mu, model.nbVar*model.nbStates, 1); reshape(s(n).Sigma, model.nbVar^2*model.nbStates, 1)];
end


%% Reproduction with GPR and DS-GMR for the task parameters used to train the model
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
disp('Reproductions with DS-GMR...');
DataIn = [1:s(1).nbData] * model.dt;
nbVarOut = model.nbVar-1;
L = [eye(nbVarOut)*model.kP, eye(nbVarOut)*model.kV];
for n=1:nbSamples
Sylvain Calinon's avatar
Sylvain Calinon committed
112
	%Rebuild model parameters with GPR, see Eq. (46)
113 114 115 116 117 118 119 120 121 122 123 124 125 126
	vOut = GPR(model.DataIn, model.DataOut, s(n).DataIn);
	
	%Re-arrange GPR output as GMM parameters
	r(n).Mu = reshape(vOut(1:model.nbVar*model.nbStates), model.nbVar, model.nbStates);
	r(n).Sigma = reshape(vOut(model.nbVar*model.nbStates+1:end), model.nbVar, model.nbVar, model.nbStates);
	r(n).Priors = model.Priors;
	r(n).nbStates = model.nbStates;
	
	%Regularization of Sigma
	for i=1:model.nbStates
		[V,D] = eig(r(n).Sigma(:,:,i));
		r(n).Sigma(:,:,i) = V * max(D,5E-4) * V';
	end
	
Sylvain Calinon's avatar
Sylvain Calinon committed
127 128
	%Retrieval of attractor path through GMR, see Eq. (17)-(19)
	currTar = GMR(r(n), DataIn, 1, [2:model.nbVar]); 
129 130 131 132 133 134
	
	%Motion retrieval with spring-damper system
	x = s(n).p(1).b(2:model.nbVar);
	dx = zeros(nbVarOut,1);
	for t=1:s(n).nbData
		%Compute acceleration, velocity and position
Sylvain Calinon's avatar
Sylvain Calinon committed
135
		ddx =  -L * [x-currTar(:,t); dx]; 
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
		dx = dx + ddx * model.dt;
		x = x + dx * model.dt;
		r(n).Data(:,t) = x;
	end
end


%% Reproduction with GPR and DS-GMR for new task parameters
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
disp('New reproductions with DS-GMR...');
for n=1:nbRepros
	%Random generation of new task parameters
	for m=1:model.nbFrames
		id=ceil(rand(2,1)*nbSamples);
		w=rand(2); w=w/sum(w);
		rnew(n).p(m).b = s(id(1)).p(m).b * w(1) + s(id(2)).p(m).b * w(2);
		rnew(n).p(m).A = s(id(1)).p(m).A * w(1) + s(id(2)).p(m).A * w(2);
	end
	%Query point vector (position and orientation of the two objects)
	rnew(n).DataIn = [rnew(n).p(1).b(2:3); rnew(n).p(1).A(2:3,3); rnew(n).p(2).b(2:3); rnew(n).p(2).A(2:3,3)];
	
	%Rebuild model parameters with GPR
	vOut = GPR(model.DataIn, model.DataOut, rnew(n).DataIn);
	
	%Re-arrange GPR output as GMM parameters
	rnew(n).Mu = reshape(vOut(1:model.nbVar*model.nbStates), model.nbVar, model.nbStates);
	rnew(n).Sigma = reshape(vOut(model.nbVar*model.nbStates+1:end), model.nbVar, model.nbVar, model.nbStates);
	rnew(n).Priors = model.Priors;
	rnew(n).nbStates = model.nbStates;
	
	%Regularization of Sigma
	for i=1:model.nbStates
		[V,D] = eig(rnew(n).Sigma(:,:,i));
		rnew(n).Sigma(:,:,i) = V * max(D,1E-3) * V';
	end
	
Sylvain Calinon's avatar
Sylvain Calinon committed
172 173
	%Retrieval of attractor path through GMR, see Eq. (17)-(19)
	currTar = GMR(rnew(n), DataIn, 1, [2:model.nbVar]); 
174 175 176 177 178 179
	
	%Motion retrieval with spring-damper system
	x = rnew(n).p(1).b(2:model.nbVar);
	dx = zeros(nbVarOut,1);
	for t=1:nbD
		%Compute acceleration, velocity and position
Sylvain Calinon's avatar
Sylvain Calinon committed
180
		ddx =  -L * [x-currTar(:,t); dx]; 
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
		dx = dx + ddx * model.dt;
		x = x + dx * model.dt;
		rnew(n).Data(:,t) = x;
	end
end


%% Plots
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
figure('position',[20,50,1300,500]);
xx = round(linspace(1,64,nbSamples));
clrmap = colormap('jet');
clrmap = min(clrmap(xx,:),.95);
limAxes = [-1.2 0.8 -1.1 0.9];
colPegs = [[.9,.5,.9];[.5,.9,.5]];

%DEMOS
subplot(1,3,1); hold on; box on; title('Demonstrations');
for n=1:nbSamples
	%Plot frames
	for m=1:model.nbFrames
		plot([s(n).p(m).b(2) s(n).p(m).b(2)+s(n).p(m).A(2,3)], [s(n).p(m).b(3) s(n).p(m).b(3)+s(n).p(m).A(3,3)], '-','linewidth',6,'color',colPegs(m,:));
		plot(s(n).p(m).b(2), s(n).p(m).b(3),'.','markersize',30,'color',colPegs(m,:)-[.05,.05,.05]);
	end
	%Plot trajectories
	plot(s(n).Data0(2,1), s(n).Data0(3,1),'.','markersize',12,'color',clrmap(n,:));
	plot(s(n).Data0(2,:), s(n).Data0(3,:),'-','linewidth',1.5,'color',clrmap(n,:));
end
for n=1:nbSamples
	%Plot Gaussians
	plotGMM(s(n).Mu(2:3,:), s(n).Sigma(2:3,2:3,:), [.5 .5 .5],.6);
end
axis(limAxes); axis square; set(gca,'xtick',[],'ytick',[]);

%REPROS
subplot(1,3,2); hold on; box on; title('Reproductions with GPR');
for n=1:nbSamples
	%Plot frames
	for m=1:model.nbFrames
		plot([s(n).p(m).b(2) s(n).p(m).b(2)+s(n).p(m).A(2,3)], [s(n).p(m).b(3) s(n).p(m).b(3)+s(n).p(m).A(3,3)], '-','linewidth',6,'color',colPegs(m,:));
		plot(s(n).p(m).b(2), s(n).p(m).b(3),'.','markersize',30,'color',colPegs(m,:)-[.05,.05,.05]);
	end
end
for n=1:nbSamples
	%Plot trajectories
	plot(r(n).Data(1,1), r(n).Data(2,1),'.','markersize',12,'color',clrmap(n,:));
	plot(r(n).Data(1,:), r(n).Data(2,:),'-','linewidth',1.5,'color',clrmap(n,:));
end
for n=1:nbSamples
	%Plot Gaussians
	plotGMM(r(n).Mu(2:3,:), r(n).Sigma(2:3,2:3,:), [.5 .5 .5],.6);
end
axis(limAxes); axis square; set(gca,'xtick',[],'ytick',[]);

%NEW REPROS
subplot(1,3,3); hold on; box on; title('New reproductions with GPR');
for n=1:nbRepros
	%Plot frames
	for m=1:model.nbFrames
		plot([rnew(n).p(m).b(2) rnew(n).p(m).b(2)+rnew(n).p(m).A(2,3)], [rnew(n).p(m).b(3) rnew(n).p(m).b(3)+rnew(n).p(m).A(3,3)], '-','linewidth',6,'color',colPegs(m,:));
		plot(rnew(n).p(m).b(2), rnew(n).p(m).b(3),'.','markersize',30,'color',colPegs(m,:)-[.05,.05,.05]);
	end
end
for n=1:nbRepros
	%Plot trajectories
	plot(rnew(n).Data(1,1), rnew(n).Data(2,1),'.','markersize',12,'color',[.2 .2 .2]);
	plot(rnew(n).Data(1,:), rnew(n).Data(2,:),'-','linewidth',1.5,'color',[.2 .2 .2]);
end
for n=1:nbRepros
	%Plot Gaussians
	plotGMM(rnew(n).Mu(2:3,:), rnew(n).Sigma(2:3,2:3,:), [.5 .5 .5],.6);
end
axis(limAxes); axis square; set(gca,'xtick',[],'ytick',[]);

%print('-dpng','graphs/demo_GPR01.png');
%pause;
%close all;