AI:Links
Skirmish AI's, are what the Computer Opponents in games are called. Developing a good Skirmish AI is a major undertaking, and requires quite a bit of know how. As the SY's have more important things to do than develop an AI, they have left this job up to the community.
The community has risen to the challenge remarkably well, releasing quite a few AI's.
There are currently 4 major skirmish AI's, AAI (by submarine), NTai(by AF), OTAI(by veylon), and KAI(by krogothe). KAI is not currently available to the public. There is also talk of Epic AI which will be the future successor of NTai and promises along with KAI to dwarf the existing AI's.
Some people may have JCAI, which was bundled with spring but is out of date. AAI 0.63 and NTai XE4 are bundled with spring atm....
Current AI's
AAI
AAI is an AI currently being developed by Submarine. AAI 0.63 is 99% stable and works with most popular mods. It also supports water and air units.
JCAI
JCAI is an AI that was being developed by Zaphod. As Zaphod has now taken over the job as head programmer, development on JCAI has come to a halt.
KAI
KAIis an AI currently being developed by krogothe. Not publicly available although krogothe has hinted at exotic ideas an algorithms, and promsies to be a powerful Skirmish AI.
NTAI
NTAI is an AI currently being developed by AF. Touting a huge featureset, it claims to be the best Skirmish AI available, taking aspects of the other AI's, and packaging itself with an array of groupAI's/HelperAI's and othertools.
OTAI
OTAI is an AI being developed by Veylon. Originally intended to emulate the Skirmish AI in OTA, it's grown from there as veylon has attempted to improve upon what the original cavedog team made.
ZcAIn
ZcAIn is an AI currently being developed by Cain. It isnt being actively developed as far as anyone is aware
Random enemies
There is a also the Random enemies script, which can be used for some basic testing. It is not an AI as such, but a spawn script that can be either used to test an AI, or to train against.
Group AIs
These AIs are designed for use by a human player, to be assigned to a group of units and give that group orders to reduce micromanaging.
Metal Maker AI
The Metal Maker AI turns member metal makers off when energy reserves get low and turns them back on again when energy reserves are nearly full.
Simple Formation AI
Central Build AI
The Central Build AI organizes the build tasks given to construction units so that units build whatever tasks are nearest to them, or guards other construction units to help them if there aren't any tasks it can build on its own.
Documentation
greenai
The greenai page has basic documentation for the Global AI interface.
Global AI Interface
Jouninkomiko has a topic on the Global AI Interface running here.
Metal Handler Class
Cain has developed and released a Metal Handler Class that succesfully finds metal spots on maps, which can be a major stumbling block for many AI developers. Information about it can be found in this thread.
Since then a new algorithm has been written by krogothe that greatly improves upon this,a nd is used by all 4 major skirmish AI's
AI:MAID
Theory
Useful existing documentation on the Spring unit AI can be found in K-man's Spring AI documentation, however this is not related to the Global AI.
Alantai Firestar has a lot of theory on the topic on his site, including details on his NTAI project.
Triaxx had a topic about the '5 Matrix AI 1,2,3', and also some information at Alantai Firestar's site.
Relevant research
- Seminar course on "Topics on AI and Computer Game Programming": The lecture slides give a great overview of advances in game AI, with plenty on RTS games
- Forbus et al 2001, "How qualitative spatial reasoning can improve strategy game AIs"
- D. Fetter (project report), "Real-Time Strategy Game Decision Making with Support Vector Machines": A project which tried using SVMs to learn unit production decisions in Warcraft III, based on replays of top players.
- R. Laursen & D. Nielsen 2005, "Investigating small scale combat situations in real time strategy computer games": 164-page Master's thesis on using game trees to determine near-optimal unit behaviors in combat scenarios (unit placement and target selection), which was implemented in the Stratagus warcraft clone. This also gives an overview of the AI techniques used by many popular games.
- N. Imrei, "Reinforcement Learning in Real-Time Strategy Games" (>final presentation link, thesis available by email): Honours project used reinforcement learning to develop both overall strategies and agent behaviors in RTS games
- S. de Jong, P. Spronck & N. Roos 2005, "Requirements for resource management game AI": This is about how to do resource management in games like Sim City, but it should also be applicable to TA-style games. There's also a 77-page Master's thesis
- Michael Buro's lab: Much of his lab's research is focused on "real-time planning and learning AI with applications to RTS games." They've actually created an open-source RTS called ORTS, in order to experiment with various AI techniques. With any luck, they might even be persuaded to switch their focus to TA Spring, or perhaps try implementing some of their algorithms. Also, it might be worthwhile to use elements of ORTS in TA Spring.
- M. Chung, M. Buro, and J. Schaeffer 2005, "Monte Carlo Planning in RTS Games"
- M. Buro 2004, "Call for AI Research in RTS Games": Gives a nice overview of the current state of AIs in RTS games
- M. Buro and T. Furtak, "RTS Games and Real-Time AI Research"
- L. MacDonald 2005, a final-project report on Moving Target Search (MTS). Basically, MTS seems to be an extension of something called Learning Real-Time A* (LRTA*).
- N. Mehta et al 2005, "Transfer in Variable-Reward Hierarchical Reinforcement Learning": Develops reinforcement-learning techniques and tests them in a simple RTS
- Forum posts
- Post by Gabba describing "Research about Pathfinding, AI and Voronoi diagrams"
- Post by NeuronExMachina on Academic research papers on AI in RTS games
