Modeling approaches, page 2

Within the biologically relevant parameter space, simple reflexive models are sometimes able to locate the odor source, but even the most successful models comprising layered control systems and centrally generated behavior fall far short of the performance of real moths. Most recently, a genetic algorithm has been employed to optimize the performance of the models. So far the genetic algorithm has identified at least two unique combinations of variable settings that result in similar success rates arising from different looking behavior.

A different and complementary approach to modeling this behavior has been taken by Dr. Danny Grünbaum, a collaborator working jointly with Drs. Belanger and Willis, in the Department of Zoology at the University of Washington. Dr. Grünbaum's approach has been to implement an analytical model of how an organism might adapt it's behavior to the ever-changing nature of an odor plume in order to track it to it's source. In this approach he has not explicitly used mechanisms known from moths (or any other living organism) to build his model. Rather, the models are organized around simple ideas, or hypotheses, for how an organism might track a chemical plume to the source. Different versions of these models can then be run in competition with one another to determine which works the best given different ecological constraints (e.g., under what conditions might "fast and sloppy" beat "slow and careful" plume tracking).

Simulating the Moth

Dr. Grünbaum's experiments have already resulted in the generation of "behavior" that we do not typically see in our simulation experiments. That is, the Bayesian moth models narrow their tracks as they approach the odor source. This so called "homing in" is often observed in the behavior of real male moths as they approach a pheromone source, but is almost never observed as output from our other simulation model. This suggests that real moths may also be using this or a similar mechanism to locate odor sources. It also suggests that including some sort of modulation of behavior based on experience may increase the success rate of our other simulation model.

One important goal of these modeling studies is to test the "rules for navigation" developed in the models by using them as the "brains" of a small robot. A new and exciting collaboration with Dr. Wayne Jouse in Aerospace and Mechanical Engineering here at the U. of A. promises to accelerate progress in the robot experiments and take them in new directions.

For our initial robotic experiments we have chosen to purchase small commercially available research robots called khepera from a company called k-team in Switzerland. These robots are designed for ease of use and are almost the same size as an adult Manduca sexta moth.

<<Previous   |   Next>>