Domain
Illustration
Methods
Conclusions
The Artificially Determined Defense is used to determine the best defensive strategy for a computer to employ versus a human opponent in a video football game. When the level of intelligence of the computer is low, the game is easily beaten so the desire to continue playing quickly diminshes. A more sophisticated strategy for the computers play selection can vastly increase the difficulty level of the game. The goal of the project is to have the agent respond to real world situations as closely as possible to a human being in order to simulate actual human-vs-human competition.
ADD will handle the determination of which defense to run based on current game situations. A possible scenario in which the agent would have to respond would be:
Fourth down and long with less than two minutes to play. The offense has the ball on their own 30 and they are down by 7.
It is expected that the team will pass based on the fact that they must score a touchdown to tie the game and since a run will take valuable time off the clock. Therefore, the agent would determine that a pass prevent defense would be optimal for the current scenario. For other scenarios the agent will also take into consideration the opponents previous tendencies. For example, if the offense were in a first and long situation, ADD would query the opponents previous play selection in that situation and assume that they would act in a similar manner.
The Artificially Determined Defense uses depth first search to traverse a series of axioms relating the current game situations to the best possible defense. When the agent encounters a fact that has not yet been defined it queries the user to obtain the necessary information pertaining to the current situation. The agent is given a list of possible facts which are askable of the user. This list is as follows:
down(D) - current down {first,second,third,fourth}
yardage(Y) - current yardage {long,medium,short}
time(T) - current time {less than 2 min,more than 2 min}
fieldpos(F) - curent field position {offensive, neutral,defensive}
half(H) - current half {first,second}
score(S) - current score {winning, losing by three or less, losing by more than three}
prev(P) - previous play for current situation {pass, run, other}
From this basic set of conditions the computer can obtain a fairly accurate representation of what to expect for the upcoming play.
The ADD can become even more sophisticated by adding additional conditions to the current list. Unfortunately, the number of factors that go into a coaches true decision are almost infinite. An exact duplication of human decision making is virtually impossible, but a practical representation is obtainable through a reasonable amount of conditions.