Game Development Reference
In-Depth Information
For example, a decision to “attack Bad Dude with our gun� means we have made a
decision about which enemy to attack, which weapon to use, that we need to draw
it, aim it, fire it, and so on. All of those other actions are included in the key deci-
sion of “attack Bad Dude with our gun.�
The reason we view “attack Bad Dude with our gun� as a single decision is that
we are processing all of the components as a whole. We could have put it up against
“attack Bad Dude with our fists,� “attack Evil Dude with our gun,� or even “attack
Evil Dude with our fists.� We were not breaking down the decision into “attack Bad
Dude or Evil Dude?� or “attack with gun or fists?� While we certainly
could have
divided the quandary into two separate parts (i.e.,
who
to attack and
how
), we may
want to score the decision based on the
combination
of the criteria.
For example, if we compare the threats posed by Bad Dude and Evil Dude, we
may find that Evil Dude is more of a threat (Figure 14.2). If we compare the rela-
tive strengths of our gun and our fists, we will likely find that our gun is a more po-
tent weapon. Those two observations may lead us to attack Evil Dude with our gun
(in the library?).
FIGURE 14.2
If Bad Dude has a particular weakness to fist attacks, separating the two
decisions would not have brought us to that choice. Only by combining the factors
into a single decision algorithm would we have discovered the correct action.
If we were to combine the target and weapon decisions together into one (mys-
terious) utility equation, however, we might find that Bad Dude has a weakness for
melee attacks and that our best option (of the four combinations) would be to
attack him with our fists. By making two separate decisions and gluing them
together, we arrived at a suboptimal behavior. By combining them, we determined
the
best
choice for the situation.
