Harm Aarts

Postal Address:
Helsinki University of Technology
Lab. of Computational Engineering
P.O. Box 9203
FI-02015 HUT, FINLAND
Visiting Address:
Tekniikantie 14, Espoo
(Innopoli 2, 3rd floor, Room B315)
Tel:
+358 9 451 4810
Email:
harm [at] lce.hut.fi

Mandatory WhoAmI

I promise I will keep it short. As per September 2005 I started my masters thesis under the supervision of Harri Valpola. Before this I was studying Artifical Intelligence at the Utrecht University. Early in my studies my attention was grasped by machine learning and especially bio inspired methods, nature does a reasonable job at learning so why can't we learn the trick from biology? Unfortunately during my studies in the Netherlands little attention was given to neural networks, genetic algorithms and the like. I consider myself very lucky to be here in Finland being able to study learning methods in such detail.

Thesis Project

During my thesis project I will try to make an TCP/IP interface between Webots and Matlab. This work will provide people willing to work with Webots an oppertunity to continue to work with Matlab with which many are familiar already and in computationally more efficient than the languages Java and C++ in which algorithms otherwise should be written. Moreover by inplementing a TCP/IP interface it is now possible to distributed the load of a particular heavy algorithm over several computers.
Next to this work I attempt to integrate earlier work modelling the cerebellum by Heikki with visual input. As Heikki's model already works with Webots, this visual input would come from a couple of simulated cameras. The goal is to make the balancing task Heikki defined capable of working in a world in which visual information carries relevant information.
This I will try in two different worlds, one in which the ground on which the robot moves induces different kinds of drag. The amount of drag can be predicted by the type of texture the ground is padded with. In the other world the ground itself is uneven and instead of recognising just the texture, the system now needs to develop some kind of depth perception. I plan to use ICA and other related methods to accomplish this.

GATE

This is the library I wrote to communicate with Webots using Java TCP/IP sockets. For more information see readme file in archive.
Download:

    • Mathlab Scripts

    These scripts proved useful during my stay here.
    webots2rgb. This script turns the mess Webots returns to Matlab with a camera_get_image command into a proper RGB image.
    stripedWorld.tar.bz2. This is a collection of files making a plane constructed from alternating texture stripes to be used in Webots.
    wavyWorld.tar.bz2. This collection of files makes a world file (.wbt) in which the ground is not an even plane but follows a wavy pattern.
    Harm Aarts
    Last modified: Fri Apr 7 14:49:26 EEST 2006