Information Technology Reference
In-Depth Information
to maximize a total reward. It considers dynamic domains and interdependen-
cies (possible constraints) among tasks. Beside the greedy centralized approach
to solving such problems, approximate solutions have been proposed, e.g. algo-
rithms modelling colonies of social insects, such as SWARM-GAP [5].
Since coordination and automatic adoption of roles are also important matters
for multiagent systems, Campbell and Wu [2] provide an extensive review of
multiagent role allocation.
In this paper, this problem is addressed by using negotiation outcomes for
the allocation of tasks defined by attributes, where the process focuses on agent
utility functions regarding these attributes, without an explicit transmission of
agent potentials.
2
Description of the Proposed Model
The following subsections formalize the definition of tasks that the agents should
negotiate for, the adaptation model of the agents, directly related to the adoption
of roles and change in productivity, and present an evolutionary approach to find
a “fair”, (near-)optimal task allocation at a given time. The model can be used to
simulate a self-organizing company, where employees are encouraged to accept
tasks themselves, rather than being assigned, similar to the recommendations
provided by the agile software developing methodologies or modern management
practices.
2.1 Task Definition
The tasks are considered to be defined by a series of
p
attributes. A task has
specific values of these attributes, considered to be
complexity levels
,eachwithin
a certain domain. Let
T
be the set of tasks
T
=
{t
i
}
,
m
=
|T |
and
F
the set
of attributes or features
F
=
{c
j
}
. Then:
t
i
=
{c
1
, ..., c
p
}
,with
c
j
∈ D
j
, ∀j ∈
{
1
, ..., p}
.
Agents have
competence levels
l
a
and
p
=
|F |
associated with each attribute
j
, which de-
scribe how much knowledge an agent
a
has regarding a particular feature
j
of a
task. These levels can increase or decrease according to the agent's experience.
The complexity levels refer to only a certain moment in the evolution of technol-
ogy. As technology evolves, one can consider that the competence level naturally
decreases, unless a person keeps up with it by learning.
It is assumed that an agent has the greatest utility when the complexity level
of an attribute
c
j
approximately matches the competence level of the agent on
that attribute. The utilities of tasks are based on the valuation of individual
attributes the task is composed of. For an agent
a ∈ A
and a task
t ∈ T
:
⎧
⎨
−
l
a
−c
j
μ
a
,l
a
− μ
a
≤ c
j
≤ l
a
1
p
u
a
(
c
j
)with
u
a
(
c
j
)=
u
a
(
t
)=
−
c
j
−l
a
ν
a
(1)
1
,l
a
≤ c
j
≤ l
a
+
ν
a
0
,
otherwise
⎩
j
=1
