Robotics Reference
In-Depth Information
is removed from the current goal list because the system can be certain
that that goal will be satisfied during the execution of the plan. There
will then be one goal less to achieve during the remainder of the plan.
If an action was selected rather than a goal, the system determines
what time it will be and what will be the state of the robot's world after
this action has been executed. If the time is too late for a deadline or if
the state of the robot's world is not to its liking, some part of the plan
may be faulty. If the system finds this to be the case it will be because one
or more goals that should already have been satisfied have not been, in
which case these goals are added back in to the partial plan to which the
action belonged, which is also updated by removing the executed action
from its list. This re-planning process is an integral part of planning and
execution, since these unsatisfied goals will require additional planning
activity, and the subsequent actions that depended on them will not be
executable until this additional planning has been carried out.
Once an action has been proposed for execution in a plan, the ro-
bot's motivations are updated in order to reflect the changes in its world.
Also the planning system's clock is updated to reflect the amount of time
taken to execute the action. The system then returns to the start of the
algorithm and continues until an entire plan has been created and found
to work within the allotted time.
Commonsense Reasoning
In 1984 Doug Lenat began work on a new approach in automatic rea-
soning, an incredibly ambitious and courageous research project aimed
at creating a massive database of real-world information that could serve
as the basis for commonsense reasoning. When it started, the project,
called Cyc, was a very long-term, high-risk gamble, which in 1994 led
to the incorporation of Cycorp, Inc., the company that markets the sys-
tem. During the first 20 years of the project, many tens of millions of
dollars were invested by U.S. Government funding agencies and others,
and more than 1.5 million pieces of knowledge were acquired.
14
The un-
derlying idea was to employ computer inferencing methods on the mass
of knowledge being collected, in order to provide the computer with
14
The figure of 1.5 million dates from early 2002. It increases as more knowledge is added but it
is reduced as similar items of knowledge within the system are merged.
