Search Unity

What do you desire most with your game bots/AI?

Discussion in 'Game Design' started by Deleted User, May 3, 2018.

  1. Deleted User

    Deleted User

    Guest

    Being a DIY guy, I am thinking about creating an AI framework for my game and I was curious to know what kinds of things other devs desire for their game AI/bots.

    What kind of behaviors come to mind when you wish to put bots in your games?
    How much customization is desired when you do this?
    Any methods of navigation/tactics/strategy you prefer?

    Trying to get a good idea of how to proceed. Things to research and work on.
     
  2. TonyLi

    TonyLi

    Joined:
    Apr 10, 2012
    Posts:
    12,706
    Skim the forum threads for other AI frameworks like Behavior Designer, Emerald AI, and ICE. It's a rich trove of pain points and beloved features. It runs the gamut, depending on the type of game people are making. Some want to run hundreds of simultaneous agents -- or that might be their first instinct, but if there were a way to LOD the AI and suspend AI that are far from the player, they may realize they'd prefer that for performance. Others want advanced behavior, from advanced squad tactics to herds and breeding.

    My primaries are:

    1. Situational awareness. Notice if you're standing next to a flaming barrel that's about to explode.

    2. Don't be dumb. Even if tactics aren't clever, at least don't do things like getting stuck in a loop that breaks the illusion.
     
  3. EternalAmbiguity

    EternalAmbiguity

    Joined:
    Dec 27, 2014
    Posts:
    3,144
    The obvious answer is that it depends on what the game calls for. I don't have a universal standard.

    For my "Dwarf Fortress in 3D" game idea, I want intelligent agents which can observe their environment around them and understand it. They should have human-like needs, along with certain personality-based desires and goals. They'll need to execute these goals in real time along with dozens of other agents, so there will be lots of interaction between them.

    For my "Sonic Adventure Redux" game idea, the enemy characters scattered around levels only need to be able to recognize when a player is in their field of view or activation area, and respond to that by attacking.
     
    theANMATOR2b likes this.
  4. newjerseyrunner

    newjerseyrunner

    Joined:
    Jul 20, 2017
    Posts:
    966
    It depends on the type of gameplay. I want AI in a platformer to be very simple and deterministic: green turtles walk straight until hitting a wall, red turtles do the same but don't walk off cliffs, and yellow turtles chase you. Always the same behavior, every single time given the same input.

    For FPS, I expect some more challenge. I like something that pushes the limits of the Turing test. It should have situational awareness, but also have long-term planning. Halo has always been had very well-liked AI, and they wrote a presentation about how they achieved it a long time ago: http://slideplayer.com/slide/1406867/ It's all smoke and mirrors, but feel like they are thinking.


    Personally, I've also experimented with neural networks. I made a dogfighting game once and I couldn't figure out how to make the AI both be evasive and fly properly so I bred a neural network with a genetic algorithm. It produced a wickedly smart AI that behaved very organically. Neural networks can be expensive to run though, so I tend to use them only for bosses or something.
     
    theANMATOR2b and Deleted User like this.