Here is a simple example that illustrates the need for sensing (and contingent plans)
We have a patient who might or might not have a disease d.
If the patient has disease, then medicating her (M) will cure her.
If however, she *doesn't* have the disease, medicating her will kill her.
There is a simple injection I that if given to a patient with the disease, makes their tongue
red; it has no effect on the patients without disease.
Your goal is to ensure that the patient doesn't have the disease (and is alive!).
The initial configuration is { D or ~D}
after injection
{(Disease, tongue-red) (disease, tongue-not-red)}
now we do a "look-see" and medicate only if the tongue is red.
So the full plan is:
Inject
If tongue-red
then medicate
{else do nothing}
Notice that this solution is not a "line" (or "sequence") plan. It corresponds to a
subgraph in the state-space.
This example also illustrates another interesting point. The "disease" is not directly
observable, but it can be indirectly observed from the tongue-redness, after the
"injection" action. (So, while sensing can reduce uncertainty, it may not be as simple as
"just find out what state you are in"--you may have to do additional world-changing actions
before you can sense).
Rao
Tuesday, January 27, 2009
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.