Li Jun and Tom Duckett

Learning Robot Behaviours with Self-Organizing Maps and Radial Basis Function Networks


Abstract

We investigate a neural network architecture for learning behaviours on a mobile robot by fusion of unsupervised and supervised learning paradigms. Specifically, a self-organizing map (SOM) for unsupervised learning and a radial basis function network (RBF) for supervised learning are integrated into one system for acquiring and performing behaviours such as wall following, obstacle avoidance and path learning. In our experimental work on a real robot, the learning system generates the non-linear mapping from the robot's sonar and infrared sensors to motor control commands, which are used to drive the mobile robot in a previously unknown environment. We discuss and analyze our preliminary results by comparing the performance of this learning system with that of a simple, linear feedforward neural net.

Download

[ps.gz] [pdf]

Bibtex 

@INPROCEEDINGS{Li02,
 AUTHOR = "Li {Jun} and Tom {Duckett}",
 TITLE = "Learning Robot Behaviours with Self-Organizing Maps and Radial Basis Function Networks",
 BOOKTITLE = "Proceedings of the Second Swedish Workshop on Autonomous Robotics",
 ADDRESS = "Stockholm, Sweden",
 YEAR = "October 10-11, 2002",
}