||Division of Inteligent Systems
Main research topics
To be able to act in an unknown environment mobile robots have to know
where they are, i.e. they have to locate themselves, using some map.
Then, they have to plan the their further path, avoiding collisions
with stationary or moving obstacles. All this is summarized under the term
Mobile Robot Navigation.
The research in the division concerns:
- 2D and 3D map building - based on sensor (laser and omnicamera) readings,
the map of an environment is built. The hybrid grid-based and object-based
representations are used.
- Localization - robots determine their position using particle filters approach.
- Path plannig - in order to plan collision free paths, Cellular Neural Networks (CNN)
are used. CNN allows combining advantages of potential field methods
and the diffusion methods.
- Coordinating the motion of a team of robots - the path for a team of robots
is planned using CNN.
Conventional CAD-tools allow the user to produce technical drawings, to visualize
the designed object in 3D-space and to perform geometrical transformations of such object.
The next generation of these tools should enable the designer to look for innovative
solutions at the conceptual phase of design, to generate efficiently the detailed
design after the principal decisions have been made, to check whether this design
fulfills all requirements stated by the Code of Practice and by the investor,
and to investigate alternative solutions.
In order to achieve such functionality, new software tools should act as expert
systems possessing certain domain knowledge and able to perform automatic reasoning.
The development of such software, usually referred to as KBD-tools (Knowledge-Based
Design tools), is on the agenda of many research teams nowadays.
The DIS takes part in this effort confining itself to the specific sub-area,
namely, to the linguistic approach to design. Treating primitives as letters
we can compose words that correspond to certain parts of the designed object
and incorporating those words into phrases we accomplish the synthesis
of the whole artifact. Introducing a grammar we can force our generative system
to produce designs that follow certain rules. On the other hand, allowing
the crossover and random mutation in the grammar rules, we can generate
The field of machine learning is concerned with the task of designing
algorithms that allow computer programs to learn automatically from
the experience. In particular, we concentrate on automatic learning
of Bayesian networks from data. Bayesian networks, also known as probabilistic
networks or Bayesian belief networks, allow a representation of joint probability
distributions in a compact way and have become popular in the field
of Artificial Intelligence. Particularly challenging is the problem
of learning Bayesian networks with hidden variables (i.e., variables
that are never observed in a given data). Our recent work concerns learning
of a latent class (also known as a naive Bayes with a hidden class variable)
and hierarchical latent class models, which are among the simplest types
of Bayesian networks with hidden variables for categorical data.
We also investigate the possibility of applying such models in robotics.
The research in the division concerns:
or, as some still prefer to say, diagrammatic representation and reasoning,
concerns the use of diagrams in information processing and communication by humans
and computers. Diagrammatic representation uses diagrams to represent
data and knowledge, while diagrammatic reasoning uses
direct manipulation and inspection of a diagram as the primary
means of inference. Diagrams are a visual kind
of analogical knowledge representation mechanism
that is characterized by a direct (though not necessarily isomorphic)
correspondence between the structure of the representation
and the structure of the represented. For more, see for example:
- Problems of errors in diagrammatic reasoning and how to avoid them
(Z. Kulpa: Self-consistency, imprecision, and impossible cases
in diagrammatic representation.
Machine GRAPHICS & VISION 12(1): 147-160, 2003).
- Development and applications of a diagrammatic representational system for
and computation (see especially the book: Z. Kulpa.
Analysis with Applications. IPPT PAN Reports 1/2006, xvi+232 pp., Warsaw 2006.
- Application of diagrams in mathematics (Z. Kulpa. On diagrammatic representation of mathematical knowledge. In: Mathematical Knowledge Management. LNCS 3119,
Springer, Berlin 2004, 190-204), especially:
- Problems of formalization of diagrammatic reasoning (work in progress).
For other publications in this area see the sections on
and on intervals in Z. Kulpa's list of publications.
Department of Intelligent Technologies
Our home page
Maintained by Michal Gnatowski
Designed by Zenon Kulpa
Last updated Sept 17, 2007