[MUD-Dev] Better Combat (long)

[David Kennerly]

My comments could be summarized as: Where is your knight?

Paul Schwanz wrote:

> How does chess manage this?  It seems to me that, although the
> rules of chess are simple enough to be summed up in one page of
> text, the complexity comes from every move being predicated upon
> all the preceding moves, both yours and your opponent's.

There are no numbers in Chess.  Only spatial relationships.  Okay,
one could put numbers into chess, and use numerical analysis.  But
it's estimated there are about 10^100 vertices on the Chess game
tree.  So get back to me when you solve Chess.  :)

> To play chess well often requires thinking ahead many moves into
> the future and correctly predicting how your opponent will react
> to your play.  However, there are those (I am not a particularly
> skilled player) who are capable of bypassing the typical
> multi-level analysis of every possible iteration and making good
> moves based on experience, intuition, and a keen understanding of
> the positional aspect of chess.

Just a hunch, but I suspect these intuitive players are delegating
the branch-and-bound game tree search to their subconscious.  The
player is still looking several moves ahead, unless you don't
consider subconscious activity to be a player activity.  This is not
to say the player searches the game tree like a computer searches a
game tree; just to say that the human and computer player are both,
as best fits their respective architecture and "software", looking
ahead--and/or pattern matching the game vertex in the tree to a
knowledge base of known vertices with known values.  I'd love to
hear from a chess programmer or a self-conscious chess player about
the internals of either intelligence.  How do chess programs think?
How do chess players think?

> I'd like to explore one in which stance, sword-point location,
> orientation to oponent, and weapon reach describe a position (not
> unlike the arrangement of chess pieces describe a position) from
> which a limited number of learned combat maneuvers (forms) are
> possible.

Thank you for posting mechanics to comment on.  It's many more
combinations than some notes on fencing I'd scribbled during the
middle of the week.

You wrote a lot about orthogonal combinations.  A broad comment
comes to mind, and that is that combinatorial size does not
necessarily increase strategic depth.  Playing tictactoe on a
3x3x3x3 grid or playing checkers with a 16x16 board is still not as
strategically engaging as playing chess on an 8x8 grid.
Combinations, alone, are not strategically engaging.  If all the
Chessmen were pawns (and remaind pawns at the end of their march) it
would be dull, even on a 10x10x10x10x10x10 board.

D&D has plenty of combinations.  They don't have stances, but they
have weapon-reach, material, size category, critical range, critical
multiplier, base damage, single-handed, and more.  But it's not
strategically interesting.  For example, critical range, critical
multiplier, base damage, and two-handed/one-handed, can all be
reduced to one attribute: damage rate.  Likewise, all D&D armor
attributes can be reduced to defense rate.  These in turn can be
further reduced to a single attribute: combat efficiency, with a
little arithmetic from relative hit points is further reduced into a
single attribute: lifespan ratio.  There are, for the statistical
sticklers, measures of variance that matter, but at some point it
still becomes a normal distribution, and so all these (except weapon
reach) are reduced from their umgumpteen-illion combinations to a
mean and a standard deviation.

That's not to say D&D is all reducible in that manner.  Weapon
reach, for example makes a strategic difference, because one can
then stand farther away.  As well, several of the feats, combat
options, spells, and *groan* attack of opportunity, add strategic
options.

For an example of MMP reduction, EQ delay and damage is reduced to damage
ratio, which Allakhazam does;

  http://everquest.allakhazam.com/itemhelp.html

So, one thing unposted and much more important than the number of
stances and reaches, is the interaction of these.  If they are all
tradeoffs of efficiencies on scales of offense/defense, then they
are reducible.  For examples:

> Armor will determine the damage done by a landed blow.

Reducible.

>    Some forms might be considered gambits.  In other words, even
>    when fully trained, they may have only a 75%, 50%, or 25%
>    chance of success, depending upon the level of gambitness (no,
>    I don't think that's a real word).

Reducible.

>    Waning endurance will result in negative modifiers to the skill
>    checks for your forms.

Reducible.

>    Forms need only be concerned with my own character's movements.
>    We can let the computer calculate the effectiveness of the form
>    based on opponent's stance, orientation, response, etc., but as
>    far as creating actual content, we only need to describe the
>    forms, not every possible position/situation.

Computable; ergo, reducible.

To echo your Chess allusion, to create a non-reducible, and
therefore potentially strategic, option you need a "knight."
Something that operates on a data structure with a pattern instead
of a simple repetition of a constant direction.  The bitch is that
it must be carefully chosen in the context of the entire system in
order to prune the game tree to bonsai-perfection that a player's
sophisticated intelligence demands.

Despite this being text-only, I imagined all the lucid descriptions,
except one:

>    Weapon Reach: But what if we move out of 2D and into 3D?
>    Imagine using a 3x3x3 grid instead to describe both your
>    character's and his opponent's stances.

Did I miss something?  Your crouch/upright qualifier already
transformed the 3x3 grid into a 3x3x2 grid.  Just a minor comment; I
still understand the conceptual design.

By the way, if this were a graphic game, by the way, that several
times more animations than the largest game I've heard of.  You
could argue for procedural animation.  At the current leaps and
bounds of graphics, could the techonology support real-time
rendering about 1000 animation poses with several times as many
transition animations sometime this decade?  I don't know the state
of the art of graphics programming.

So, I'm a little unclear on what the significance of different
combinations are.  If it is a spatial position, then there is
clearly a distinct set of optimal stances, hand positions, etc. for
each one being encountered.  I like how you manage to define

The way you defined weapon reach, it sounded like a weapon that was
longer penetrated both shallower and deeper into the personal space
of the target.  Since it thus threatens more cubes in the 3-grid, a
longer weapon dominates a shorter one.  that Did I overlook some
qualitative or quantitative countermeasure?

In many fighting games, there are distinct blocks and attacks that
have distinct counters.  Many players find this fun.  A huge genre
surrounds it.  Yet in these, the sets are crisp.  There are clearly
good and bad responses, with very few mediocre responses.  It's
usually hit or miss, with no gentle slope of intermediate options.

A gentle (continous, monotonic, strictly increasing) fitness slope
is trivial for any evolutionary algorithm to climb.  Likewise, such
fitness slopes are reducible to the old binary search guessing (pick
a number between 1 and 100...).  Thus, dull for a sophisticated
player, after discovering this property.  In addition to collision
detection, are there other /qualtitative/ differences?  This would,
analogous to a knight's move, is neither forward nor aside, nor
continuously through, but a discrete leap.  In any case, thank you
for the tour.

Sincerely curious,

David
_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev