Robotics Reference
In-Depth Information
methodology of GPS, growing a tree of possibilities in the search for a
solution.
In common with many other types of search problem in AI, planning
gives rise to enormous tree growth, as programs are required to search
vast problem spaces. During the 1990s, the technology of automatic
planning proliferated, producing several powerful techniques that over-
come the problems presented by vast search spaces. As planning meth-
ods grew more sophisticated, the types of problem with which planner
systems were faced grew more intricate. Not only were planners being re-
quired to cope with relatively straightforward problems, those that could
be solved completely on a step-by-step basis
before
the first step was ac-
tually undertaken, new types of problem were attracting interest in the
world of AI, problems that involved more than one robot and problems
that required certain steps to be taken before the later steps in the plan
could even be considered.
Ruth Aylett and her team at the University of Salford have developed
a software design for a robot that must plan continuously, as for exam-
ple the robots in a soccer team.
12
In her design, the robot's goals are
generated as a result of its motivations. Having autonomy in the goals
it selects makes the robot's planning considerably more sophisticated—
the planning process interleaves planning steps with execution steps and
sometimes a planning step can involve delegating a task to a different
robot.
Aylett's software architecture is designed to provide continuous plan-
ning for a robot by meeting the following requirements:
1. Planning and executing are interleaved—the robot plans a step and
then examines its situation as it will be if that step is executed.
2. The planner is able to accept new goals generated by the robot at
any time.
3. Time passes while the robot plans and executes, so the planner
must be able to reason about time.
4. Actions are planned for many robots so the planner must be able
to reason about which robots should carry out which actions.
5. It may not always be possible to achieve all current goals within the
given time, so it must be possible to prioritise outstanding goals. It
12
See the section “Robot Soccer” in Chapter 8.
