In the Movable Feast Machine (MFM), a simulated indefinitely scalable computer, we have to contend with a highly limited scope of memory and random asynchronous execution, which complicates search algorithms and data structures.

We took on the problem of finding other elements in a sparsely populated system using an active sensing system. In our simulation, we have a number of ant colonies which have to find and destroy each other. The scouts from each colony leave behind 'pheromone trails', which act like linked lists, and can be “activated” to lead other ants to enemies.

We found that the wandering-scout model, combined with the self-optimizing trails, made for a highly effective means of discovering routes and building a network organically in the MFM.

This method for pathfinding works well for sparse structures, demonstrating both an ability for the concept to work with a clear goal in the context of a more complex system (the conflict) and an ideal density level for the system to allow pathfinding to work well. If the machine is too sparsely populated, it can take scouts far too long to find their adversaries, while in a densely populated system the optimization (and eventually, the movement of the scouts) is severely limited. Additionally, when breadcrumbs persist too long, they can contribute to system clutter; I found that giving them a built-in maximum age of [X] helped mitigate this, as well as causing them to die after they are finished functioning for a single alert cycle.