Talks@IDSIA: Tim Taylor and Varun Raj

Istituto Dalle Molle di studi sull’intelligenza artificiale

Data d'inizio: 15 Dicembre 2009

Data di fine: 16 Dicembre 2009

IDSIA is pleased to announce the following talks:
11:30 Dr Tim Taylor
12:00 Dr Varun Raj
Both talls will be held on Tuesday, the 15th of December,  in Sala Primavera, Galleria 2 bldg, 6928 Manno. Abstracts follow.
Engineering open-ended evolution
Tim Taylor

One of the grand challenges in artificial life is to understand how to build systems capable of indefinitely continuing evolutionary change,
which exhibit a comparable capacity for complexity, diversity and creativity as observed in their natural counterparts.  The ability to
create artificial systems with this kind of rich, on-going evolutionary potential (i.e. "open-ended evolutionary systems") would
have profound implications for the automated production of complex artifacts, artificial life and artificial intelligence.  However, all
attempts at creating this kind of system to date, including systems of self-replicating computer programs (e.g. Tom Ray's Tierra, or my own
Cosmos), together with attempts at evolving real RNA molecules in vitro, have met with failure; they quickly reach a quasi-stable state
beyond which no further qualitative changes are observed.

In earlier work I analysed the reasons for these failures, and proposed ways in which they may be overcome.  One of the key problems
is in designing a system in which agents can come to exploit their environments, and to interact with each other, in novel and creative
ways beyond those which have already be "programmed into" the system.
To solve this problem involves looking at it from a different angle, and modelling agents and environment as a single dynamical system.

I will describe results from some initial pilot studies of this new approach, which exhibit, among other things, the emergence of agents
that can solve tasks by evolving new forms of sensing and control of their environment that were not specifically "programmed in".  I will
end the talk by discussing a number of issues that arise from this new modelling perspective, which define the focus of my ongoing research in this area. 

Image-based Detection of Semi-transparent Objects
Varun Raj, Honda Research Institute Europe
Most computer and robot vision algorithms, be it for object detection, recognition, or reconstruction, are designed for opaque objects. Non-opaque objects have received less attention, although various special cases have been the subject of re- search efforts, especially the case of specular objects. The main objective of this thesis is to provide a seminal work in the case of semi-transparent objects, i.e. ob- jects that are transparent but also redirect light, typically objects made of glass. They are rather omnipresent in man-made environments (especially, windows and doors). Detection of these objects provides vital information that can be used in a robot's localization and path planning. Also, several other important applications are dis- cussed in the report. In order to achieve the detection of semi-transparent objects we developed a novel approach using a collective-reward based technique on an image captured by an uncalibrated camera. We also present a robotic-vision version of the approach in the form of a semi-transparent obstacle avoidance algorithm for a wheeled mobile robot. Several experiments were conducted over several scenarios to test the efficacy of the algorithm.