One thing to keep in mind is that Gladiator/Q3A bots followed very simple patterns and never played like a human would. Over the time, modders made their own bots that were much more human-like.
Spiterbot was a notorious one. It was capable of strafejumping, rocketjumping, zone denial, pickup timing, risk/reward behavior, enemy power estimates, and all things you would expect from a competent human player. Things like pro-q3dm6 bridge-to-railgun jump weren't a problem for it. It was still too easy to read for a good player, and susceptible to mistakes leading to HP/armor starvation, like avoiding the railgun control line like plague where a human would have tried to outsmart an enemy, risked it, or waited for an enemy mistake. Too bad there never was a technical write-up for the Spiterbot like this one.
(also, the article is from 2001, this should probably be in the title)
- navigation, which is done with a combination of A* or Dijkstra and navmeshes
- decision, which is done with the paradigm of behavior trees
- computer vision, to complete what navmeshes can not provide
Despite the term "computer vision", it is still not AI.
I will quote the "Navigation" and "Computer vision" section from the site I linked above:
> All bots should be able to reach any part of the map that does not require special training.
> Bots should be able to detect any enemy in their FoV. It would be nice to have bots not being too good at finding hidden buildables on the roof. This makes hiding eggs unfun and next to impossible in the current state. Unless they are placed outside of a navmesh, in which case bots are totally unable to discover them.
One thing to keep in mind is that Gladiator/Q3A bots followed very simple patterns and never played like a human would. Over the time, modders made their own bots that were much more human-like.
Spiterbot was a notorious one. It was capable of strafejumping, rocketjumping, zone denial, pickup timing, risk/reward behavior, enemy power estimates, and all things you would expect from a competent human player. Things like pro-q3dm6 bridge-to-railgun jump weren't a problem for it. It was still too easy to read for a good player, and susceptible to mistakes leading to HP/armor starvation, like avoiding the railgun control line like plague where a human would have tried to outsmart an enemy, risked it, or waited for an enemy mistake. Too bad there never was a technical write-up for the Spiterbot like this one.
(also, the article is from 2001, this should probably be in the title)
We had bots in Tremulous which is an old ioquake3 fork, there was no AI involved, worked fine. See https://github.com/cmf028/CoW-Tremulous.
Unvanquished (https://unvanquished.net) became the new "Tremulous" (different engine, etc.), also has bots.
Read more about it here: https://wiki.unvanquished.net/wiki/Bot_design
- navigation, which is done with a combination of A* or Dijkstra and navmeshes
- decision, which is done with the paradigm of behavior trees
- computer vision, to complete what navmeshes can not provide
Despite the term "computer vision", it is still not AI.
I will quote the "Navigation" and "Computer vision" section from the site I linked above:
> All bots should be able to reach any part of the map that does not require special training.
> Bots should be able to detect any enemy in their FoV. It would be nice to have bots not being too good at finding hidden buildables on the roof. This makes hiding eggs unfun and next to impossible in the current state. Unless they are placed outside of a navmesh, in which case bots are totally unable to discover them.
There is a pull request that implements the use of "beacons" for bots which is used to locate buildables: https://github.com/Unvanquished/Unvanquished/pull/2683
There was no "computer vision" in the Tremulous bots, FWIW.
In Tremulous, they used waypoints only, so has nothing to do with AI either.
Bots in Unvanquished are quite human-like and there are plans to add personalities (if they did not add them already) and there is no AI there either.
TL;DR: You can have bots that work quite damn well and human-like without AI/ML.
Awful paper. Almost zero citations!
> The ultimate goal of AI is to create an artificial man
Close tab.
To expand on this: https://news.ycombinator.com/item?id=45924721
No AI required whatsoever!