Robotics Reference
In-Depth Information
A very simple evaluation function for Chess, a function with only
one feature, would add up the values of the pieces on each side and
assign a score corresponding to the total net value. But Chess is much
more complicated than that, and besides, in many situations a player
must choose between two or more moves that lead to positions with the
same material balance, so material itself is normally not sufficient as a
distinguishing factor.
Another important heuristic employed by human Chess players is
“mobilize your pieces”, and the second most important feature in Chess,
after material, is the relative mobilities of the two armies. Other things
being equal, the player with the more mobile army has the advantage.
In Chess we measure mobility by the number of legal moves that can be
made by all the pieces in a position. Simply add up how many moves
can be made by each white pawn, knight, bishop, rook, queen and the
king, and the total is a reasonable measure of White's mobility. Carry out
the same calculation for Black and the difference in mobility scores will
generally indicate which player has the advantage and by how much.
With two features in its evaluation function, material and mobility, a
Chess program will need numerical weights to indicate the relative im-
portance of each feature. Material is much more important than mobility
so its weight must be large enough to reflect this. A ratio of between 5-
to-1 and 10-to-1 would probably be appropriate here, meaning that an
advantage is mobility of somewhere between five and ten moves is equiv-
alent to an advantage in material of one pawn. The exact ratio chosen for
the weighting of each feature will affect the “style” in which the program
plays, by making it more likely or less likely to be willing to sacrifice a
pawn here and there in return for a significant advantage in mobility.
After material and mobility there are several other features of a Chess
position that often contribute to a player's advantage, for example con-
trol of the central squares. This can be measured by assigning different
squares on the board different values, according to how near or far they
are from the centre. It is possible to create an evaluation function for
Chess with almost any desired level of sophistication. Bonuses can be
given for “developing” one's pieces early in the game (i.e., bringing them
out from their original squares so they are taking some part in the ac-
tion), for attacking the area around the opponent's king, for protecting
one's own king, and so on. The more sophisticated the evaluation func-
tion becomes, the more accurate will be its numerical estimate of the
merit of a position. But there is also a down side—the more complex
