demo_TPGMR_LQR01.m 6.98 KB
Newer Older
Milad Malekzadeh's avatar
Milad Malekzadeh committed
1
function demo_TPGMR_LQR01
Milad Malekzadeh's avatar
Milad Malekzadeh committed
2 3 4 5 6 7
% Demonstration a task-parameterized probabilistic model encoding movements in the form of virtual spring-damper
% systems acting in multiple frames of reference. Each candidate coordinate system observes a set of
% demonstrations from its own perspective, by extracting an attractor path whose variations depend on the
% relevance of the frame through the task. This information is exploited to generate a new attractor path
% corresponding to new situations (new positions and orientation of the frames), while the predicted covariances
% are exploited by a linear quadratic regulator (LQR) to estimate the stiffness and damping feedback terms of
Milad Malekzadeh's avatar
Milad Malekzadeh committed
8 9 10
% the spring-damper systems, resulting in a minimal intervention control strategy.
%
% This demo presents the results for a finite horizon LQR.
Milad Malekzadeh's avatar
Milad Malekzadeh committed
11
%
Milad Malekzadeh's avatar
Milad Malekzadeh committed
12 13 14
% Author:	Sylvain Calinon, 2014
%         http://programming-by-demonstration.org/SylvainCalinon
%
Milad Malekzadeh's avatar
Milad Malekzadeh committed
15 16
% This source code is given for free! In exchange, I would be grateful if you cite
% the following reference in any academic publication that uses this code or part of it:
Milad Malekzadeh's avatar
Milad Malekzadeh committed
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
%
% @inproceedings{Calinon14ICRA,
%   author="Calinon, S. and Bruno, D. and Caldwell, D. G.",
%   title="A task-parameterized probabilistic model with minimal intervention control",
%   booktitle="Proc. {IEEE} Intl Conf. on Robotics and Automation ({ICRA})",
%   year="2014",
%   month="May-June",
%   address="Hong Kong, China",
%   pages="3339--3344"
% }

%% 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)
Milad Malekzadeh's avatar
Milad Malekzadeh committed
33
model.dt = 0.01; %Time step
Milad Malekzadeh's avatar
Milad Malekzadeh committed
34 35 36 37 38 39 40
nbRepros = 8; %Number of reproductions with new situations randomly generated
rFactor = 1E-1; %Weighting term for the minimization of control commands in LQR


%% Load 3rd order tensor data
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
disp('Load 3rd order tensor data...');
Milad Malekzadeh's avatar
Milad Malekzadeh committed
41 42 43 44
% 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
Milad Malekzadeh's avatar
Milad Malekzadeh committed
45 46 47 48 49 50 51 52
% 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 (5).
load('data/DataLQR01.mat');


%% Tensor GMM learning
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
fprintf('Parameters estimation of tensor GMM with EM:');
Milad Malekzadeh's avatar
Milad Malekzadeh committed
53
model = init_tensorGMM_timeBased(Data, model); %Initialization
Milad Malekzadeh's avatar
Milad Malekzadeh committed
54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70
model = EM_tensorGMM(Data, model);


%% Reproduction with LQR for the task parameters used to train the model
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
disp('Reproductions with LQR...');
DataIn = [1:s(1).nbData] * model.dt;
for n=1:nbSamples
	%Retrieval of attractor path through task-parameterized GMR
	a(n) = estimateAttractorPath(DataIn, model, s(n));
	r(n) = reproduction_LQR_finiteHorizon(DataIn, model, a(n), a(n).currTar(:,1), rFactor);
end


%% Reproduction with LQR for new task parameters
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
disp('New reproductions with LQR...');
Milad Malekzadeh's avatar
Milad Malekzadeh committed
71
for n=1:nbRepros
Milad Malekzadeh's avatar
Milad Malekzadeh committed
72
	for m=1:model.nbFrames
Milad Malekzadeh's avatar
Milad Malekzadeh committed
73
		%Random generation of new task parameters
Milad Malekzadeh's avatar
Milad Malekzadeh committed
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
		id=ceil(rand(2,1)*nbSamples);
		w=rand(2); w=w/sum(w);
		rTmp.p(m).b = s(id(1)).p(m).b * w(1) + s(id(2)).p(m).b * w(2);
		rTmp.p(m).A = s(id(1)).p(m).A * w(1) + s(id(2)).p(m).A * w(2);
	end
	%Retrieval of attractor path through task-parameterized GMR
	anew(n) = estimateAttractorPath(DataIn, model, rTmp);
	rnew(n) = reproduction_LQR_finiteHorizon(DataIn, model, anew(n), anew(n).currTar(:,1), rFactor);
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
axis(limAxes); axis square; set(gca,'xtick',[],'ytick',[]);

%REPROS
subplot(1,3,2); hold on; box on; title('Reproductions with finite horizon LQR');
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 Gaussians
	plotGMM(r(n).Mu(2:3,:,1), r(n).Sigma(2:3,2:3,:,1), [.7 .7 .7]);
end
for n=1:nbSamples
	%Plot trajectories
	plot(r(n).Data(2,1), r(n).Data(3,1),'.','markersize',12,'color',clrmap(n,:));
	plot(r(n).Data(2,:), r(n).Data(3,:),'-','linewidth',1.5,'color',clrmap(n,:));
end
axis(limAxes); axis square; set(gca,'xtick',[],'ytick',[]);

%NEW REPROS
subplot(1,3,3); hold on; box on; title('New reproductions with finite horizon LQR');
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
	%Plot Gaussians
	plotGMM(rnew(n).Mu(2:3,:,1), rnew(n).Sigma(2:3,2:3,:,1), [.7 .7 .7]);
end
for n=1:nbRepros
	%Plot trajectories
	plot(rnew(n).Data(2,1), rnew(n).Data(3,1),'.','markersize',12,'color',[.2 .2 .2]);
	plot(rnew(n).Data(2,:), rnew(n).Data(3,:),'-','linewidth',1.5,'color',[.2 .2 .2]);
end
axis(limAxes); axis square; set(gca,'xtick',[],'ytick',[]);

%print('-dpng','outTest1.png');

Milad Malekzadeh's avatar
Milad Malekzadeh committed
146 147
%Plot additional information
figure;
Milad Malekzadeh's avatar
Milad Malekzadeh committed
148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
%Plot norm of control commands
subplot(1,2,1); hold on;
for n=1:nbRepros
	plot(DataIn, rnew(n).ddxNorm, 'k-', 'linewidth', 2);
end
xlabel('t'); ylabel('|ddx|');
%Plot strength of the stiffness term
subplot(1,2,2); hold on;
for n=1:nbRepros
	plot(DataIn, rnew(n).kpDet, 'k-', 'linewidth', 2);
end
xlabel('t'); ylabel('|Kp|');

%Plot accelerations due to feedback and feedforward terms
figure; hold on;
n=1; k=1;
plot(r(n).FB(k,:),'r-','linewidth',2);
plot(r(n).FF(k,:),'b-','linewidth',2);
legend('ddx feedback','ddx feedforward');
xlabel('t'); ylabel(['ddx_' num2str(k)]);

%print('-dpng','outTest2.png');
%pause;
%close all;