Sign in to follow this  

What AI Problems do YOU want discussed at the AI Summit?

This topic is 561 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

How an AI manages a large-scale battle in an RPG with many units and unit types. For example - When to retreat, which units should target others, group movement during the battle, consideration of reinforcements.

Share this post


Link to post
Share on other sites

I do have a question, and it's a big one, leaving your panel gasping for air :p

It's not game-related, but may be fun to ask anyway.

 

As you may know, RoboCup is a world-wide effort to create intelligent robots. http://www.robocup.org/about-robocup/

RoboCup@home is a part of it, where the aim is to build robots that act intelligently in real-time, interacting with humans in a house or public area. RoboCupRescue aims to build robots to find and help people in disaster locations.

 

A real-world environment is a highly confusing world. A house is a very dis-organized place. Tasks are very diverse, and perhaps not even known beforehand. Real-time interaction puts heavy constraints on what can be computed. In closer interaction (eg lifting someone out of bed, or digging someone out after a house collapsed), safety is a big concern.

 

What can the (game) AI community advice to make this work?

Share this post


Link to post
Share on other sites

Maybe to abstract but I would love to get insight in how a professional would approach SRPG/TRPG AI namely how to get good team work rather than just separate entities all just doing there own thing

 

That's actually a decent (and relatively common) one. Phrase it as something specific and I'm sure we can make it work.

Share this post


Link to post
Share on other sites

I do have a question, and it's a big one, leaving your panel gasping for air tongue.png

It's not game-related, but may be fun to ask anyway.

 

As you may know, RoboCup is a world-wide effort to create intelligent robots. http://www.robocup.org/about-robocup/

RoboCup@home is a part of it, where the aim is to build robots that act intelligently in real-time, interacting with humans in a house or public area. RoboCupRescue aims to build robots to find and help people in disaster locations.

 

A real-world environment is a highly confusing world. A house is a very dis-organized place. Tasks are very diverse, and perhaps not even known beforehand. Real-time interaction puts heavy constraints on what can be computed. In closer interaction (eg lifting someone out of bed, or digging someone out after a house collapsed), safety is a big concern.

 

What can the (game) AI community advice to make this work?

 

Way too broad and only loosely related to game dev. Remember, our audience is going to be game AI devs who are looking for solutions they can apply to their own projects. Perhaps by turning this on its head and simply asking something along the lines of handling cluttered, unknown, PCG, environments? Of course, we already have a session about that this year courtesy of Damian Isla and John Abercrombie.

 

http://schedule.gdconf.com/session/ai-for-generated-worlds 

Share this post


Link to post
Share on other sites

 

How an AI manages a large-scale battle in an RPG with many units and unit types. For example - When to retreat, which units should target others, group movement during the battle, consideration of reinforcements.


Spoilers.

 

 

Hmmm... this keeps coming up. Odd, that...

Share this post


Link to post
Share on other sites

I don't know.how about:

 

How to get good team work out of individual entities in an SRPG/TRPG.

or 

How professionals would approach SRPG/TRPG AI.

or

Putting the teamwork in Tactical game AI.

or

Tactical game AI a professional approach.

 

As you can tell I'm not particularly good at this, anyone want to help me come up with a title.

Share this post


Link to post
Share on other sites

I'm looking more for a game-specific situation. "Say I have a group of agents in an RPG that each have their own abilities. How do I get them to use their abilities the best way, but yet also be aware that their teammates might also be better-equipped or better-positioned to do something than they are?"

Share this post


Link to post
Share on other sites

Actually I have a second related question:  Lets say I have an TRPG with 3 characters able to attack an enemy in the same turn.  Now lets say one character has the possibility to stun the enemy and lets say one has low hp.  It would make sense to have the character with stun attack first to potentially minimize the damage done to the other characters assuming the character with the stun is not of low HP.  So not only a matter of what the best move is but the best order.  I suppose you could extend this thinking even further and what the optimal combined attack with minimal loss would be.  Or something like that.

Share this post


Link to post
Share on other sites

Actually I have a second related question:  Lets say I have an TRPG with 3 characters able to attack an enemy in the same turn.  Now lets say one character has the possibility to stun the enemy and lets say one has low hp.  It would make sense to have the character with stun attack first to potentially minimize the damage done to the other characters assuming the character with the stun is not of low HP.  So not only a matter of what the best move is but the best order.  I suppose you could extend this thinking even further and what the optimal combined attack with minimal loss would be.  Or something like that.

You could easily do this with a min-max algorithm. In fact, it would be the expected outcome from applying an appropriate min-max.

Share this post


Link to post
Share on other sites

 

Actually I have a second related question:  Lets say I have an TRPG with 3 characters able to attack an enemy in the same turn.  Now lets say one character has the possibility to stun the enemy and lets say one has low hp.  It would make sense to have the character with stun attack first to potentially minimize the damage done to the other characters assuming the character with the stun is not of low HP.  So not only a matter of what the best move is but the best order.  I suppose you could extend this thinking even further and what the optimal combined attack with minimal loss would be.  Or something like that.

You could easily do this with a min-max algorithm. In fact, it would be the expected outcome from applying an appropriate min-max.

 

 

Actually, setting up a min-max algo with 3 characters on the same side would be a pain. MCTS would work better, I suppose, but when you are trying to figure out the sequencing stuff, a planner might actually do a bit better.

 

Of course, there are ways of setting up a utility system with a blackboard that allows teammates to register what they could possibly do and use a priority weighting structure to sort out the options.

Share this post


Link to post
Share on other sites

Actually, setting up a min-max algo with 3 characters on the same side would be a pain. MCTS would work better, I suppose, but when you are trying to figure out the sequencing stuff, a planner might actually do a bit better.

I really don't see how a MCTS would be better than min-max? 3 characters per team is a pretty small search space. It seems easy to cut the tree down with a simple evaluation function (Enemies dying = Very good / Team members dying = Very bad / Enemies lose health = Good / Team members lose health = Bad). You could go some moves deep with brute force (whatever is computationally reasonable) before pruning begins.
Anyway, yeah, I don't see how this is much different from implementations of mix-max elsewhere or why it would be a pain, or how MCTS would fair better?

(sorry for going a little off-topic, I just would like to see why MCTS would be preferable with such a small search space with an evaluation function looking pretty easy to write too)

Edited by Mats1

Share this post


Link to post
Share on other sites

Actually, setting up a min-max algo with 3 characters on the same side would be a pain. MCTS would work better, I suppose, but when you are trying to figure out the sequencing stuff, a planner might actually do a bit better.

I really don't see how a MCTS would be better than min-max? 3 characters per team is a pretty small search space. It seems easy to cut the tree down with a simple evaluation function (Enemies dying = Very good / Team members dying = Very bad / Enemies lose health = Good / Team members lose health = Bad). You could go some moves deep with brute force (whatever is computationally reasonable) before pruning begins.
Anyway, yeah, I don't see how this is much different from implementations of mix-max elsewhere or why it would be a pain, or how MCTS would fair better?

(sorry for going a little off-topic, I just would like to see why MCTS would be preferable with such a small search space with an evaluation function looking pretty easy to write too)


I am also inclined to suggest MCTS here, not because of the size of the search space, but because MCTS applies to a much larger class of games than minimax. For instance, MCTS works well even if the results of an attack have a random component; or if there are more than two sides in the conflict, each with its own objectives. Edited by Álvaro

Share this post


Link to post
Share on other sites

What order would you be putting your plys in? In standard minmax, it is "me, opponent, me, opponent...". When you are trying to sort out which "me" is going first, the complexity is largely inside each ply itself. Even just the ordering of 3 characters leads to a few different states to solve, i.e. ABC, ACB, BAC, BCA, CAB, CBA. Now if you add on the possibility that each of those characters could have different types of actions, you get a whopper of a combinatorial explosion. Just in one ply of your min-max.

Share this post


Link to post
Share on other sites

A quest generator in a somewhat contained environment such as a farm (as one entity in a much larger world)  by where as various different needs/improvements occur quests/tasks are generated, including the ability to react to unexpected player behaviours such a chicken massacre resulting in the farm generating a quest/task to replenish chicken supplies or quests/tasks not performed by players lead to additional issues and what becomes a priority in order to restore to working order versus an utter collapse/reduced operation of the "farm".

 

 

Sorry for the layman's language - I haven't dealt with coding in what seem to be centuries.

Edited by Stormynature

Share this post


Link to post
Share on other sites
What about AI directors, how would you approach them ?

Example:
You have a bunch of individual agents, everyone with his own set of abilities, subdivided into two groups. These groups encounter each other dynamically and a combat situation emerge.

I would like to see a AI director detecting such a scene and starting to direct the combat situation to make it more visual appealing. So, instead of each agent just following its own concerns (attack highest danger first, hide from counter etc.), a director would overwrite the behavior to generate a more interesting situation,e.g. a policeman using his radio to call support, two thugs in the backline dashing forward to interrupt him, while a police commander calls orders to someone else to protect the one with the radio....

Share this post


Link to post
Share on other sites

I would say  :  How to get domain knowledge INTO a game AI system.    Building of the logic which the AI runs for the particular task (whatever aspects of the game the AI is run for).

 

This would be largely offline, though part of such logic might be recognizing what ways a player plays and then adjusting/applying appropriate strategies/tactics against it.

 

The bulk of logic for even simple AI systems is prohibitive, so methods of (partially) automating it to improve/increase  the amount a game can possess.

Edited by wodinoneeye

Share this post


Link to post
Share on other sites

This topic is 561 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this