Abstract
In this paper, the pheromone-oriented behavior of moths will be demonstrated by synthesis with biosensors and a small mobile robot that is controlled by recurrent neural networks. Since antennae on a silkworm moth are very sensitive as compared to conventional arti ficial gas sensors, they can be used as living gas sensors that detect pheromone molecules. A simple recurrent artificial neural network was used to control pheromone-tracing behavior in the manner of a living male silkworm moth. This neural network generates mothlike behavior while interacting with the environment. The turning behav ior, in particular, is a suitable tactic for small intelligence when a robot misses pheromone molecules. Our neural network is so simple that it can be very easily used as the controlling devices for micro- robots, which have a small amount of space for intelligence. Our robot is a hybrid system that combines living organisms and artificial machines, and is therefore a new type of robot.
This approach has the advantage of a real-world experiment with biosensors instead of computer simulation. The difference between the real world and the simulation conditions yields a discrepancy in results. An additional discrepancy is generated by the sensor model. Therefore, real-world experiments with living antennae may provide a fascinating interface between computer simulation and neuroethology.
Get full access to this article
View all access options for this article.