Ant Colony–Based Simulation 125
appropria te number of ants to each task is necessary to increa se the fitness of
the colony as a whole. Ants can divide labor without any centralized control;
such autonomous distribution of tasks would be useful in the field of rob otics or
scheduling in factories. The next equation is proposed as a model for assigning
workers to multiple ta sks through distributed control.
T
θ
(s) =
s
n
s
n
+ θ
n
(5.13)
For instance, T is defined as tasks such as feeding larvae. The probability that
a given ant does this task T
θ
is determined by the amount of pheromone s
that the larvae emit and the thre shold for each individual θ.
In reality, larvae secrete more pheromone when they are hungry, and reduce
the amount of pheromone secretion when caretakers p e rform their tasks. In-
dividual ants go out to collect food when the amount of detected pheromone
becomes lower than a threshold value, and conversely, when the amount of
pheromone from larvae exceeds a threshold, ants that returned from collect-
ing food become caretakers. There is a distribution of the threshold in each
individual; therefore appropriate numbers of individuals can be distributed to
multiple tasks.
Such a behavior model can be applied to the problem of task distribution
in robots capable of multiple tasks or fault-tolerant systems. For example, a
solution that uses agents called routing wasps has been proposed for scheduling
tasks in a factory [16]. In this system, pseudo-pheromones ar e emitted from
tasks in a queue based on priority and wait time. Agents are assigned to each
assembly machine, and thresholds to perform specific tasks are determined
based on the status of each machine. Agents a ssign tasks to each machine
with probabilities determined by threshold values and amounts of pheromone.
Such a system was shown to increase throughput of a factory.
Ant methods are being implemented in various ways in industry. For ex-
ample, Bios Group
1
based in New Mexico is a consulting firm which builds
systems based on s warm intelligence, and has provided methods to make
scheduling efficient to Southwest Airlines and P&G, for example. P&G uses
distributed scheduling where collaborative decisions such as transport of raw
materials and management of factories are made by agents on a network. The
swarm approach is used to build a system where the transport path is deter-
mined by taking into account the utilization of overcrowded warehouses in the
candidate paths.
1
http://www.biosgroup.com/