Game Development Reference
In-Depth Information
with, such as being convex. Path finding within a node in the navigation mesh is
done locally, and simple methods often suffice; if the node is convex, then a
straight line is all that is required.
An issue with any pathfinder is that long paths are fragile paths. Bridges get
blown up, freeways are subject to being closed by accidents, and fly-over rights
can be denied after a flight takes off. Before blindly using A* and expecting all
paths to work, you should investigate the methods that are being used to deal
with path problems.
Before the novice AI programmer rejoices that path finding is solved, he or she
needs to consider the needs of the virtual world. A* will rightly determine that
using roads and bridges is better than taking cross-country treks and fording
streams. What A* does not do is traffic control. Unexpected traffic jams on
popular routes and near choke points are realistic, but so are traffic cops. The
game's AI might be smart enough to prevent gridlock, but the planning code still
has a problem. The planning code has to factor in the effects of the very traffic the
planning code is dispatching. If the planning code places a farm tractor, a truck
carrying troops, and a motorcycle dispatch rider on the same narrow path in that
order, the whole convoy is slowed to the speed of the tractor. A* told the planner
that the dispatch rider will get to its destination in far less time than will actually
happen. A* told the planner that the troops will get there on time as well. Even
worse is the case when opposing traffic is dispatched to both sides of a one-lane
bridge. A* by itself may not be enough.
Machine Learning
Machine learning has been ''the next big thing'' in game AI for many years. The
methods have been described as elegant algorithms searching for a problem to
solve. The algorithms have been put to good use in other areas of computer
science, but they have gained very little traction in game AI. People have been
looking for ways of getting around the limits of machine learning in games for
some time now [Kirby04].
Realizing the potential of machine learning is difficult. There are many kinds of
machine learning. Two of them are popular discussion topics: neural networks
and genetic algorithms. Neural networks appear to be very enticing, particularly
to novice game AI programmers. They have been a popular topic for over a
decade in the AI roundtables at the Game Developers Conference. The topic is
Search Nedrilad ::

Custom Search