So far, Robotic Operating System (ROS) is used mainly in the field of robotic research. We want to change this and introduce this modular system also into the Artificial Intelligence.
The upcoming ROS library should provide reusable sub-systems for architectures of artificial agents. The nodes provided here will be freely available and should provide implementations of selected algorithms useful in AI and ALife domain. The resulting nodes will also be usable by simulator of hybrid architectures: Nengo+ROS.
You will be welcome to download these nodes and use them by yourself.
The main focus will be on these features:
- Unsupervised learning
- Online learning
- Domain independence
First Sub-systems in Progress
Our first sub-systems (nodes) were created as part of Bachelor and Diploma theses by our students. We hope that these nodes will be available to download and use very soon. These are now:
Pattern and Sequence Recognition module
This work should provide two ROS nodes:
- Pattern recognition module
- Sequence recognition module
The nodes were created as part of Bachelor Thesis of Pavol Sekereš and should be capable of online unsupervised learning.
Self-Organising Map and Planning module
This Bachelor Thesis written by Lukáš Skála builds hybrid architecture for solving Block World problem with noisy and uncertain input data. The architecture uses:
- Self-Organising Map (SOM) for simplification of input data
- STRIPS planning algorithm for planning the actions towards solution
Unsupervised Approximation of Dynamic Systems
Another important feature of artificial agents operating in real-word, dynamically changing environment is its ability to learn dynamic properties of particular events. This Bachelor Thesis is written by Martin Paňko and should provide subsystem wit ability of unsupervised approximation of observed dynamics.