mainpage.c

遗传算法和神经网络结合

/* This is a dummy c file that includes mainpage documentation */

/**  
\mainpage YAKS - Yet Another Khepera Simulator

\section intro Introduction
Like many other research groups we have for some years now used 
<a href="http://www.k-team.com">Khepera robots</a> in our 
research on robot learning and evolution using artificial neural networks (ANN). A number of 
Khepera simulation tools are now available and commonly used for this type of research, each 
of which has its respective advantages and disadvantages: 
<a href="http://www.cyberbotics.com">WEBOTS</a>, for example, a 
commercial product of Cyberbotics SA, offers powerful graphical 3D visualization and analysis 
tools, but unfortunately it is not freely available. Olivier Michel's Khepera Simulator, 
WEBOTS' 2D-predecessor, is freely available and has been used for some of the robot learning 
work in our group (Z96, Z96, L99, L00, LJ01), but it has the disadvantage that experiments can 
take quite much time, in particular in the case of evolutionary robotics experiments, due to the 
fact that the graphics, at least in the standard version, cannot be switched off.
Stefano Nolfi's Evorobot is also freely available and it comes with a user-friendly 
Windows-based graphical user interface (GUI), a choice of built-in standard ANN architectures 
and a built-in evolutionary algorithm, which allows the user to quickly implement own 
experiments and replicate those of others. Evorobot and its predecessor kepsim, which we 
have used in some of our evolutionary robotics work (CZ99, Z99, Z00, JHZ01), have the 
disadvantages, from our perspective, that (a) for historical reasons the ANN code and the robot 
simulator code are tightly integrated, and (b) the built-in ANN architectures are limited to the 
most common feed-forward and recurrent multi-layer perceptron network architectures and 
variations thereof. Due to the fact that much of our work is concerned with novel and/or non-
standard ANN architectures, such as recurrent higher-order networks (e.g. Z99) or unsupervised 
vector quantizers (e.g. L00), we have in many cases been forced to extend the simulator's 
functionality by programming `around' the built-in code, often leading to unnecessarily 
inefficient program code. <br><br>
Hence, the motivations and objectives for developing yet another Khepera simulator, the YAKS 
system presented here, have been as follows:
<ul>
<li>
the robot simulator code and the code for ANNs and evolutionary or learning 
algorithms should be clearly separated modules, and it should be possible to use 
one without the other;
</li>
<li>
the functionality of the robot simulator part should be equivalent to that of the 
kepsim simulator (ONL95), whose source code had kindly been made available to 
use by Stefano Nolfi and Henrik Hautop Lund, and wherever possible it should be 
compatible to it, such that, e.g. robot environments could easily be exhanged 
betweeen the two simulation environments;
</li>
<li>
the simulator should offer powerful tools, comparable to those in the above simu
lators, for visualization (2D though) and analysis of robot behavior and the under
lying ANN mechanisms;
</li>
<li>
any graphics/visualization should, however, be completely optional and thus 
should not slow down the simulation of evolutionary or learning processes;
it should be possible to compile and use the simulator on different system plat
forms (UNIX, Linux, MS Windows);
</li>
<li>
both robot simulator and ANN library code and documentation should be made 
publically available on the web.
</li>
</ul>

  
\section install Installation
  
\subsection step1 Step 1: Building the simulator
There are two type of targets that can be build, the simulator without
any graphics which gives a fast stable simulator that are ideally to 
be running long evolutions but not so plausible for testing and evalution.
Therefore we have the option to build a variant with a graphical user 
interface, ideally created for evaluation and testing of fitness function
and such alike. We distuingish the two variants by prefixing the target
with a g if we want a GUI.

So if we presumes that you have set up the Makefile to match your system
configuration we can simply build the simulator with GUI by typing
make gsim in the root of the source directory. This will build the
system and put libraries in ./lib and target binaries in ./bin the default
name for the target with GUI is gsim.

This ofcourse gives us the target name for the simulator without GUI, 
which in this case would be sim.

So if you later want to add own targets - for sanitys sake please
try to follow this codex by prefixing the GUI target with a g.

\subsection step2 Step2: Setting up the Makefile

gsim
sim

    
etc...

*/