Bay 12 Games Forum

Please login or register.

Login with username, password and session length
Advanced search  

Author Topic: My experimental AI game for university, also a testbed for something much bigger  (Read 256 times)

Foamy

  • Bay Watcher
    • View Profile




You can get it here: Link

I've been working on this for ages, and the larger game for over a year now. Mostly completely silently while we prepared. Hunter's Moon is not as far along as I feel it should be before reveal, but the testbed needs exposure and so HM has to be there too. The main push of development is still to come where most of it will be completelt overhauled. That time is only a week or so from starting.

Here's the description of the testbed:

The plan for this game was to develop a greatly simplified version of the Hunter's Moon (see website for details) hunting mechanics that didn't depend on any of the main game features, and that put the focus on the AI.

The first reason for the testbed is to act as a research project for my final year of university by analysing the merits of training a complex Multi-Agent System with a Genetic Algorithm rather than doing it manually, the second reason is to aid the development of Hunter's Moon.

The data for analysis is provided by a user study at the end of each play-through of the game.

The game will randomly pick an AI configuration at the start of each game (either manually tuned, or computer tuned), that adjusts the behaviour of the wolves. Both attempt to maximise the fun of the game.

After this project is concluded more AI development will be done to make it do everything HM requires and to flesh it out, but the basic foundation is here.


Details on the big game, Hunter's Moon, can be found here: Link

It's based on this ancient DF mod idea: Here


Feedback and survey responses are very much appreciated.
Logged

darkflagrance

  • Bay Watcher
  • Carry on, carry on
    • View Profile

It looks like most of the work has been in melee combat and injury simulation. Can you shed more light about AI goals and behavior? I find it hard to believe that you can offer an AI experience to match the variety of scenarios you plan to offer in your prototype build. However, hunting a cunning and brutal beast who has to be outsmarted and adapts to my tactics seems to be part of the final experience you plan to offer.

Also, you have "rouge-like" appearing in your blog post.
Logged
...as if nothing really matters...
   
The Legend of Tholtig Cryptbrain: 8000 dead elves and a cyclops

Tired of going decades without goblin sieges? Try The Fortress Defense Mod

Foamy

  • Bay Watcher
    • View Profile

Which scenarios are you referring too?

The AI at the moment is only considering the pure hunting case, although the architecture would support several others without heavy modification. The prototype build will only be the hunting scenario, additional ones will come in as development continues. If we need to then we'll probably implement a hybrid AI system, using the current Multi-Agent System idea in combination with more traditional methods.

Adapting to individual tactics is something we really want to add in future, that's going to have to wait a while though. Also, there's huge potential, and few obstacles, to training the AI on a wider scale. If it can learn from one person, then if you add a little internet access you can learn from thousands.

Typo fixed, good spot.
Logged

darkflagrance

  • Bay Watcher
  • Carry on, carry on
    • View Profile

Nevermind, I'm reading your AI testbed page and it has the information I was looking for. I had assumed it was simply the download page.
Logged
...as if nothing really matters...
   
The Legend of Tholtig Cryptbrain: 8000 dead elves and a cyclops

Tired of going decades without goblin sieges? Try The Fortress Defense Mod

NRDL

  • Bay Watcher
  • I Actually Like Elves
    • View Profile
Logged
All hail Deon, creator of the Genesis Mod!

GOD DAMN IT NRDL.

Foamy

  • Bay Watcher
    • View Profile

Analysis of results is tomorrow; I have a reasonable number of results but not as many as hoped. Any additional surveys would be really useful. No need for extra comments in them either.

Thanks for all the help so far. I think this study is going to be very valuable for the future.
Logged