1
$\begingroup$

I'm working on a multi-player game that involves a board/map where there are walls, cover or open ground. The players then take turns moving units around and undertaking discrete actions, like move, fire etc.

I've been exploring adding an AI for single-player games, or to enhance multi-player games. I've looked at:

  • Behaviour Trees
  • Utility-based decisions
  • GOAP

The problem I see with all of these is that in order for the AI to use specific multi-unit tactics, like a pincer attack, or leap-frogging units; that behaviour needs to be explicitly added separate to the above.

So, how can a Behaviour Tree tell an AI how to identify it should use a Flanking manoeuvre, i.e. move a unit around to hit from the side? How does the AI identify that it's a good plan, that is action across multiple "moves"? How can the AI identify where to move the individual units in order to pull this off (how does it identify the flank)?

$\endgroup$
2
$\begingroup$

Reinforcement learning is the best fit for problems like these. To get started explore how DeepMind solved Atari games. The key is self play. RL works through trial and error and eventually figures out the best strategy. You also need to tweak so that it does not play at God level making it impossible for users to win.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.