Evolving dynamical neural networks for adaptive behavior
Adaptive Behavior
Cognition is Not Computation; Evolution is Not Optimisation
ICANN '97 Proceedings of the 7th International Conference on Artificial Neural Networks
| Hi-index | 0.00 |
The design and implementation of a possible alternative to Artificial Neural Networks (ANNs) for agent controlis described. This alternative, known as the nBrain, uses the phase-space representation as the inspiration for a controller. The agent progresses to different states due to the presence of attractors in phase space, whose locations are set by the robot's genome. Several experiments with simulated agents are described. The experiments were successful in that tasks which have been performed by ANNs in the past were successfully accomplished by nBrains under evolution. The possible advantages and disadvantages of nBrains over ANNs are discussed, and directions for future work are presented.