Page 325 - Introduction to AI Robotics

308
Multi-agents
8
“hear” the progress of others in teams, as well as their own progress. If they
get frustrated with their own progress, they should stop what they’re doing
and move on to something else. Likewise, if a robot is free and another robot
has been unable to accomplish a task, it should try to complete the unfinished
task. This is particularly useful for tasks where there is a logical sequence of
behaviors, where all of a particular task (like dusting) needs to be done for an
area before the robots begin working on another task (e.g., sweeping). These
MOTIVATION changes in behaviors are regulated by a simple mechanism: motivation.The
motivation of a robot to do a task is regulated by two internal motivations,
ROBOT IMPATIENCE robot impatience and robot acquiescence. The more frustrated a robot with gets
ROBOT ACQUIESCENCE with another robot’s performance on t i , the higher the impatience associated
with that task t i . Likewise, the more frustrated a robot gets with its own per-
formance for a task, the higher the acquiescence. If the frustration threshold
is exceeded, then the robot either takes over the unfinished task or abandons
its current task and changes behavior.
Fig. 8.6 shows the time trace for an example of motivation for two space
ants foraging for asteroids. (This example isn’t really a sequential series of
tasks in the manner used by ALLIANCE, but this conveys the elegance of
motivation.) In this case, the reactive space ants have to either broadcast
what they’re doing or be able to perceive the other’s progress. This makes it
a bit different than the “no communication” approach. At time 0, both robots
start by looking for asteroids. (We assume there is no frustration for the find
task.) Both see asteroid A1, but Robot 1 is the first there. Robot 1 has now
taken responsibility for Task 1 (T1), pushing A1 to home. Even though A1
is still stationary at time 3, Robot 2 does not join in as it would in the no-
communication method. Instead, it begins to accrue impatience about T1.
Once Robot 1 begins to push A1, it starts accruing frustration in the form of
acquiescence. As with the no-communication example, a single robot cannot
push the asteroid.
While Robot 1 is trying to push asteroid A1, Robot 2 sees and moves to
asteroid A2. All the while its impatience over T1 is growing. At time 7, Robot
2 is trying unsuccessfully to push asteroid A2 (task T2) and its acquiescence
counter is increasing. Also at time 7, Robot 2’s patience with Robot 1 and
task T1 has been exceeded. It pushes T1 onto its stack of things to do when
it completes its curent task. Meanwhile, at time 9, Robot 1 gives up on T1.
Although it is frustrated with Robot 2, it assumes that T2 is still under control
and so begins to forage again. Finally, at time 10, the frustration over T2
reaches the limit and Robot 1 is free to help Robot 2.