147 lines
6.1 KiB
Plaintext
147 lines
6.1 KiB
Plaintext
|
|
||
|
Challenge ProMode Arena: Enhanced Bots
|
||
|
--------------------------------------
|
||
|
Date: 28 Apr 03
|
||
|
|
||
|
Well, go figure...
|
||
|
Thanks to Qrealka and Diamant, I've put some effort into the bots.
|
||
|
None of this would have happened if they hadn't been willing to help
|
||
|
out, because it would have taken more time than I have to spare.
|
||
|
Pretty much all the DM AI has been improved over CPMA1.1 with new
|
||
|
decision-making code. The bots are quite a bit smarter than before,
|
||
|
and at times you can even "see" what they're thinking if you watch
|
||
|
them. They're still completely ignorant of many things though, and
|
||
|
unfortunately are still dependent on AAS files, which are invariably
|
||
|
flat out wrong about level connectivity. Swelt or I may write a guide
|
||
|
to how mappers can fix some of the worst problems if we ever get time.
|
||
|
I've ONLY focused on DM, so I expect they're still absolutely awful
|
||
|
at TDM/CTF/etc.
|
||
|
|
||
|
Fixes are generally for the new bots only, but some have been applied
|
||
|
to the standard bots as well if the bug is too pathetic to tolerate.
|
||
|
Those are marked with an 'I' (e.g. the 1.1 RL aim bug was a "fixI")
|
||
|
and are implicitly id bugs, obviously.
|
||
|
|
||
|
The skill thresholds are:
|
||
|
79- has pretty sucky aim
|
||
|
81+ enables evil RL
|
||
|
86+ tries to avoid "bad" RG shots **
|
||
|
bot_challenge 1
|
||
|
enables evil hitscan at skill 91+ after you're open for 0.5-0.25s
|
||
|
doesn't degrade accuracy at close ranges
|
||
|
aims faster after using a tele
|
||
|
does NOT otherwise affect aim skill
|
||
|
uses the "alternate" weapon choices (more hitscan)
|
||
|
|
||
|
** At 86+, you'll see a noticeable jump in hitscan accuracies.
|
||
|
This DOESN'T mean that they've suddenly become all-aim bots. They
|
||
|
actually still only hit as many shots as they do at 85, but since
|
||
|
they FIRE far less often when they know the shot is a "bad" one,
|
||
|
the percentages become much higher.
|
||
|
|
||
|
Skill level also affects movement speed to some extent.
|
||
|
|
||
|
There are some new bots as well, with slightly better strategic
|
||
|
play and some "personal" properties. They can only be used at
|
||
|
skill 6 and above (CPMA AI).
|
||
|
|
||
|
Apheleon
|
||
|
faster than all other bots
|
||
|
has improved RL
|
||
|
uses the "alternate" weapon choices
|
||
|
|
||
|
arQon
|
||
|
only slightly faster than normal bots
|
||
|
has godlike RL
|
||
|
uses RL a LOT
|
||
|
|
||
|
rat (the LPB version)
|
||
|
faster than normal bots
|
||
|
has godlike RL
|
||
|
uses RL a LOT
|
||
|
|
||
|
|
||
|
add: strafejumping, yay!
|
||
|
Qrealka and Diamant get all the credit for this one
|
||
|
|
||
|
add: tanked bots will hunt you down and spawnrape you >:)
|
||
|
|
||
|
add: evil RL prediction (quite ratlike at times :P)
|
||
|
|
||
|
add: a bit of dodging and ducking for cover
|
||
|
|
||
|
chg: skill 6-100 uses the new AI (1-5 are the id bots)
|
||
|
|
||
|
chg: rewrote item choice management (esp wrt health and armour)
|
||
|
|
||
|
chg: weapon selection now biased towards killshots
|
||
|
|
||
|
chg: removed bot_boost
|
||
|
|
||
|
fixI: bots completely ignoring enemies 3ft away from them
|
||
|
|
||
|
fixI: bots unable to hit enemies at very narrow angles
|
||
|
|
||
|
|
||
|
Date: 20 Mar 03
|
||
|
|
||
|
|
||
|
CPMA's bots have undergone several dozen bugfixes and tweaks over the past
|
||
|
few years, especially wrt melee, but have continued to use the id code for
|
||
|
the bulk of their "decision" making, which is essentially "wander around
|
||
|
at random collecting items until we blunder into someone."
|
||
|
|
||
|
I've generally been reluctant to work on the AI as I don't consider it a
|
||
|
particularly challenging or interesting area, since the vast majority of
|
||
|
its problems are buried deep within the engine itself and the BSPC/AAS
|
||
|
process, improvements generally have to be made by fighting against Q3
|
||
|
every step of the way.
|
||
|
|
||
|
However, the continually decreasing "quality" of my ISP, the staggeringly
|
||
|
inept Charter Communications, has forced my hand. You can thank their sheer
|
||
|
incompetence and near-total inability to run a network that functions at
|
||
|
all, let alone at an acceptable level, for these improvements. :/
|
||
|
|
||
|
With 1.1, the "strategic" gameplay behaviour of the AI has undergone some
|
||
|
fairly drastic changes, which is to say it actually exists now. :P
|
||
|
|
||
|
Bots are aware of armour and health, and will time them to a reasonable
|
||
|
degree. They also have a basic understanding of the relative value of
|
||
|
items, i.e. given the choice between an RL and a box of MG ammo they'll
|
||
|
usually go for the RL; and they will no longer walk within inches of a YA
|
||
|
and ignore it completely to collect a shard 20ft away instead.
|
||
|
How well this approach works is however strongly dependent on the map.
|
||
|
Since bots still use the id code for their movement, they will still play
|
||
|
fairly poorly on the vast majority of maps, including id's own. In many
|
||
|
cases, the combination of defective AAS files and limitations of the
|
||
|
movement code will result in them performing WORSE than before as they
|
||
|
continually try and fail to reach certain items.
|
||
|
The CPM maps have always been designed for the most highly-skilled human
|
||
|
players, and as such are usually something of a "worst case" environment
|
||
|
for the Q3 bots. CPM3a however (Swelt's remix of jude's classic CPM3) has
|
||
|
undergone extensive tweaking to make it as bot-friendly as possible,
|
||
|
while retaining the exact same human-gameplay as the original.
|
||
|
|
||
|
The other key bot-related change in 1.1 is a fix to some defective math in
|
||
|
their use of projectile weapons. While it's historically been true that
|
||
|
there is nothing LESS dangerous than a bot with an RL, this is no longer
|
||
|
the case. While still very much inferior to a good human player, they can
|
||
|
now be quite a threat.
|
||
|
|
||
|
Bot skill as used by g_spSkill and the addbot command now ranges from 1-15.
|
||
|
|
||
|
1-5 is "vaguely-id" AI, which is to say that it uses the id melee code but
|
||
|
otherwise doesn't really have much in common with baseq3. I'm not going to
|
||
|
back out 3 years of bugfixes.
|
||
|
|
||
|
6-10 is "standard" CPMA AI, i.e. what 1-5 was in 1.0
|
||
|
|
||
|
11-15 is "enhanced" CPMA AI, as outlined in this file. Since I've barely
|
||
|
spent a day on this, it's really only DM-aware so far. When I get time
|
||
|
I'll teach them about TDM and CTF: splitting the map into zones of
|
||
|
responsibility; sharing weapons with teammates; etc etc.
|
||
|
|
||
|
Regardless, skill 100 bots will absolutely thrash normal "Nightmare" bots,
|
||
|
and when combined with friendly maps and bot_boost 1.5 can actually be
|
||
|
decent opponents rather than mindless cannon fodder. They're fun. :)
|