startup

 Matlab toolbox for planning in POMDPs with continuous spaces.

 Matlab toolbox for the POMDP planner described in 
   Point-Based Value Iteration for Continuous POMDPs
   Journal of Machine Learning Research 7, 2006 pp. 2329-2367
 by
   Jm Porta, N. Vlassis, M.T.J. Spaan, and P. Poupart

 This code is a from-scratch re-writting of the original code used for the
 experiments reported in the paper. Several things are different in this
 this version (some minor bugs are fixed, some models are changed....) so,
 expect (small) changes in the results to be obtained with this version and
 those in the paper.

 All the experiments deal with a very simple one-dimensional path planning
 problem just to demostrate the viability of the presented approach. 

 The weaker point of the presented approach is the observation model.
 Strong assumptions are made on this model and this limits the practical
 applications of this approach, specially when the observation model is
 continuous. 
 If I even continue this research I will try to circumvent this issue.

 The main stress in this implementation is put on clarity and not in
 efficientcy. 

 The best way to understand this software package is:
   A) To read the paper.
   B) To check the Experiments directory where you can find specific
      functions to replicate the different experiments reported in the paper.
      We refer to those experiments by the figure number where their results
      are plotted. It should be relatively easy to use this toolbox on your
      own problems using these functions as a template.
 
 For details on each function, check the associated documentation in the
 Doc directory (Doc/index.html)

0001 % Matlab toolbox for planning in POMDPs with continuous spaces.
0002 %
0003 % Matlab toolbox for the POMDP planner described in
0004 %   Point-Based Value Iteration for Continuous POMDPs
0005 %   Journal of Machine Learning Research 7, 2006 pp. 2329-2367
0006 % by
0007 %   Jm Porta, N. Vlassis, M.T.J. Spaan, and P. Poupart
0008 %
0009 % This code is a from-scratch re-writting of the original code used for the
0010 % experiments reported in the paper. Several things are different in this
0011 % this version (some minor bugs are fixed, some models are changed....) so,
0012 % expect (small) changes in the results to be obtained with this version and
0013 % those in the paper.
0014 %
0015 % All the experiments deal with a very simple one-dimensional path planning
0016 % problem just to demostrate the viability of the presented approach.
0017 %
0018 % The weaker point of the presented approach is the observation model.
0019 % Strong assumptions are made on this model and this limits the practical
0020 % applications of this approach, specially when the observation model is
0021 % continuous.
0022 % If I even continue this research I will try to circumvent this issue.
0023 %
0024 % The main stress in this implementation is put on clarity and not in
0025 % efficientcy.
0026 %
0027 % The best way to understand this software package is:
0028 %   A) To read the paper.
0029 %   B) To check the Experiments directory where you can find specific
0030 %      functions to replicate the different experiments reported in the paper.
0031 %      We refer to those experiments by the figure number where their results
0032 %      are plotted. It should be relatively easy to use this toolbox on your
0033 %      own problems using these functions as a template.
0034 %
0035 % For details on each function, check the associated documentation in the
0036 % Doc directory (Doc/index.html)
0037 %
0038 clc;
0039 disp(' ');
0040 disp('                    Software associated with the paper: ');
0041 disp('            Point-Based Value Iteration for Continuous POMDPs');
0042 disp('                                   by');
0043 disp('           Jm Porta, N. Vlassis, M.T.J. Spaan, and P. Poupart');
0044 disp('     Journal of Machine Learning Research Vol. 7, pp. 2329-2367, 2006');
0045 disp(' ');
0046 disp(' ');
0047 % Add all paths of the toolbox recursively
0048 path(path,genpath(pwd));