|A Designer Diary
| Ant robots
| Argentine ants and ant Robots
I recently read The Lives of Ants by Laurent Keller and Elisabeth Gordon. It is a fascinating book, a study of the lives of ants from a perch at the cutting edge of modern science. Not a run-of-the-mill nature book displaying the usual suspects. Unless you are a professional myrmecologist, you are guaranteed a few startling moments of awe at the heart of modern science. The English translation can be a bit heavy-handed in places and the writers seem to have some difficulty reconciling academic and popular styles. Rather than lessening the impact of the book, I found it remarkable that The lives of Ants conveys the excitement of myrmecological discovery without recourse to the brilliant literary and political genius of E. O. Wilson.
Biological systems, like their non-living counterparts, lend themselves to mathematical modeling. One striking example in the book concerns the way Argentine ants "determine" which food source lies nearest to the nest. Ant scouts go out to look for food in all directions, laying pheromone trails as they go. When a scout finds a food source, she picks up a piece of food and follows her own scent trail back to the nest. Now other scouts sense her trail and start going back and forth between the food source and nest, laying more pheromone trails with each trip. The ants working on the nearest food source do more round trips, laying their pheromone trails most often. As the shortest trail is marked most often, it contain the most pheromone per unit length, making it the most "smelly". Other ants are attracted by the stronger smell and follow their sisters on the shortest trail to the food. In a short space of time, the colony would be working most actively on the nearest source of food.
Ants may call it dinner but this engineer saw a system with positive feedback. As with all systems with positive feedback, this mode of search is likely to be unstable. Eventually, more ants would want to use the short trail than the path and food heap could allow. One wonders what happens when the system reaches saturation.
Roboticists looking at ant behavior saw more than another example of positive re-inforcement. Ant robotics, a branch of swarm robotics, is the study of relatively simple robots designed to accomplish complex tasks in unpredictable environments through their mutual interaction. Strong in numbers rather than individual sophistication, ant robots lay pheromone (chemical or otherwise) trails in order to communicate simple messages about themselves and the environment to their nearest neighbors. The artificial ants are often very different from real ants in their appearance, functionality and behavioral model. Like neural networks, swarmbots and ant robots are biologically inspired rather than artificial copies of living systems. Some ant robots attempt to copy the physical behavior of ants, as in forming a "live bridge" over an obstacle by hooking their bodies together. Some use a virtual form of evolution by natural selection to dynamically adapt a robot "species" to its environment. The common thread, though, is the distribution of a complex task in a way that is not only redundant and de-centralized but truly emergent.