- Abstract:
-
Homing navigation behavior is seen in many insects and animals. It is believed that in some insects the behaviour involves calculation of path integration. In this paper, a circular neuron cell structure in which each neuron accumulates distance travelled in a particular direction is suggested as a suitable computational structure for finding a proper homing vector. The neural network integrates information about angular direction to the nest and distance from the current position to the nest, from the path in which an agent has moved. Knowledge of exterocentric direction is assumed to be based on a sensor such as a light compass. A second circular array of neurons is used to compute the homing direction. The network has the characeristics that each neuron has a simple linear activation function, except inhibition neurons connected to each direction cell activation, and that connections between cells are linearly weighted. Similar to the model proposed by Wittmann and Schwegler, our approach differs in where path components are integrated and in the read-out mechanism for the homing vector computed by the model. This neural structure may partly explain homing behaviour in insects and animals and is also potentially useful as for robotic homing systems. The proposed neural network architecture was tested using a Khepera robot simulation.
- Copyright:
- 2004 by The University of Edinburgh. All Rights Reserved
- Links To Paper
- No links available
- Bibtex format
- @InProceedings{EDI-INF-RR-0219,
- author = {
Kim DaeEun
and John Hallam
},
- title = {Neural network approach to path integration for homing navigation},
- book title = {Proceedings of From Animals to Animats 6},
- publisher = {MIT press},
- year = 2004,
- month = {Jun},
- }
|
