Information Technology Reference
In-Depth Information
following constraint : their postconditions must match the mandatory pre-
conditions of the current task. The founded tasks are executed as a general
OR-task.
2. Regulatory conditions : They are not blocking unlike the mandatory condi-
tions. But in the case of an ALT-task, if there is no tasks to execute, the
planner can try to solve the regulatory conditions.
3. Favorable expertise conditions : These conditions are interpreted only by
expert agent. If they are not true agent will not execute the task. Example
of environment conditions : (
>Pipe
6
diameter
50)
4. Favorable contextual conditions : They are not blocking. They are useful
in the case of an OR-task, if all the tasks can be executed then the condi-
tions allows to choose the more appropriate task depending on the context.
Example of contextual conditions : (=
weather state rainy
)
5. Safety conditions : Same as the favorable conditions, only expert agent will
evaluate the safety environment conditions.
6. Resources conditions : First we check if agent has the resource in its tool
box, in its pockets or in its hands. If agent does not have the resource, if
it is not in a hurry it will look for a plan to get the resource. It has also
the possibility to ask another agent. If agent is in opportunistic mode it will
reason by analogy. This means it looks for a similar tool that can be used to
do the task (a-sort-of object).
5R su s
We developed each operator as a cognitive autonomous agent on the OMAS
platform. We have an agent in charge of managing the objects database and an-
swering queries. Each operator agent has an interface agent called MIT
4
which
is in charge of sending to its interface module developed with QT all informa-
tion about the agent's states, the value of its different characteristics and its
plan. These values are use to show the evolution in real time of the agent's
states and plans. We tested our algorithms on a small part of the scenario
(Fig. 9).
Agents have to do the task ”Drain the pipe”. This order matches the goal
g
1
(=
pipe
0
6
state drained
). The planner generates a plan to achieve this goal.
Fig. 10 illustrates the action selection mechanism results for three steps. Yel-
low tasks are
active
, greens are
ended
and red are
in failure
.Atstep1,
Agent
1
chose to prepare the recuperation by putting a bucket under the
Pipe
0
6. The
other choice is a task related to safety, inexperienced agent can not do this
type of task.
Agent
2
that is expert, chose the second way that is to throw
absorbent and ammonia on the floor. The second main task of this little sce-
nario is to open the gate so that the liquid can flow. There is one of two ways
to do the task : turn the handle of the gate slowly or roughly.
Agent
1
chose
the first alternative that is to turn slowly. As the handle is rusted the task
4
Masverp InTerface.
