Hacker News new | past | comments | ask | show | jobs | submit login
Three States and a Plan: The A.I. of F.E.A.R (2006) [pdf] (media.mit.edu)
69 points by Wlad007 on Aug 6, 2016 | hide | past | favorite | 24 comments



Ahh GOAP This is a very interesting approach to AI. This topic was a huge part of my FYP. I got this to work on a GPU with cuda.

Anyways, http://alumni.media.mit.edu/~jorkin/goap.html (more from the same author : Jeff Orkin)

has a lot more information about this approach.

https://github.com/stolk/GPGOAP (not mine) for anyone who wants to look at this in action (c/c++ code).



I would be interested in knowing how does constraining state complexity in AI get handled when designing games that must meet certain framerate and performance reqs- especially in FPS games? Are there instances where decisions are forced because of the need for an agent to undertake an action? And what heuristics are used? This is definitely part of chess/ board game playing AIs with time restrictions.

Has any work been done in trying to model AI as a deep learning problem, getting testers to play against one another and running feature extractors on their behaviors (a la deep mind) and training NN models on this data?


Google's Go playing AI which recently beat some world champions? Sounds just like what you describe.


As I described in my question- board games ai aka go this has been done for years. I'm interested in fps games


I've read about half the article so far, but isnt this basically prolog?

Is there reference to efficency in the document? Constraint solvers like prolog are pretty slow.


They added a cost per action and use A* for graph traversal instead of Prolog's depth-first search.


I am not a gamer, but I did know of F.E.A.R.

The paper is from 2006? Are these still valid techniques, or have they been supplanted with evolving A.I.?


The most popular choice today are behavior trees. They are very powerful, modular, and there are even visual tools for putting your building blocks together.

The Unreal engine also comes with such a tool for example:

https://docs.unrealengine.com/latest/INT/Engine/AI/BehaviorT...

A good intro which highlights flexibility and reuse:

http://www.gamasutra.com/blogs/ChrisSimpson/20140717/221339/...

Having said that, FEAR's AI held up really well. It's still top-notch.


Thanks for the links! Glad to hear that work carries over, so I can revisit it as a good example.


Still valid and still somewhat beyond the state of the art for most genres of game. The point of an AI in a game experience typically isn't to be the best possible intelligence, but to support the experience by "losing gracefully". This inevitably leads to a very scripted/authored approach, where autonomous planning has to be curtailed.

In the worst case, overpowered game AI creates an unstable global economy that spins into chaos the moment play begins, leading to mysterious aftermath situations ("why are all these people dead?") that the player stumbles upon hours later. Some of the Bethesda games experimented with this sort of high-level autonomy and ultimately dialed it back to ship because it was too hard to avoid chaos.


Not out of disbelief but rabid curiosity, if you happen to have a link to a gamasutra article or gdc talk or something similar where bethesda folks talk about their AI experiments running amok I would love to read it.


I'd like to read it as well, but out of disbelief.

Bethesda has been marketing 'radiant AI' bullshit for their past three games and never delivering. I really doubt they ever developed anything resembling high level autonomy.

I haven't followed the development of their games extremely closely but I read a good amount of game development news/marketing and I never saw anything convincing, only non technical talk. The most I ever really remember them talking about it was in an E3 demo for Elder Scrolls: Oblivion way back when where they lied by misleading the audience to think that an obviously scripted quest had something to do with their radiant AI system.


There are a few quotes floating around the web, all with the exact same wording, but I cannot find the original source anymore! I'm not sure if it's bitrotted or if it's a story someone made up before the release of the game, and then people spread it until it became "true". But it is what it is:

> 1. One character was given a rake and the goal “rake leaves”; another was given a broom and the goal “sweep paths,” and this worked smoothly. Then they swapped the items, so that the raker was given a broom and the sweeper was given the rake. In the end, one of them killed the other so he could get the proper item.

> 2. In another test, a minotaur was given a task of protecting a unicorn. However, the Minotaur repeatedly tried to kill the unicorn because he was set to be an aggressive creature.

> 3. In one Dark Brotherhood quest, the player can meet up with a shady merchant who sells skooma, an in-game drug. During testing, the NPC would be dead when the player got to him. The reason was that NPCs from the local skooma den were trying to get their fix, did not have any money, and so were killing the merchant to get it.

> 4. While testing to confirm that the physics models for a magical item known as the “Skull of Corruption,” which creates an evil copy of the character/monster it is used on, were working properly, a tester dropped the item on the ground. An NPC immediately picked it up and used it on the player character, creating a copy of him that proceeded to kill every NPC in sight.

> 5. In one test, after a guard became hungry and left his post in search of food, the other guards followed to arrest him. The town people looted the town shops, due to lack of guards.

> Bethesda worked to fix these issues, balancing an NPC’s needs against his penchant for destruction so that the game world still functions in a usable fashion. In-game there are over 1,000 different NPCs, not including randomly spawned monsters and bandits. The result is that the AI in the release version is much reduced, only featuring NPC schedules.


I believe it. Even in the released game, the NPCs' schedules break down on certain edge cases and they revert back to this kind of chaotic behavior. I've watched Quill-Weave sneak across town and follow someone into a temple to steal food out of his pockets, because her travel schedule (regularly visiting a certain town) was disrupted by interactions with the player.


> have they been supplanted with evolving A.I.?

Not to my knowledge. In fact, the AI of F.E.A.R. is still more or less looked at as a benchmark of good FPS AI. I'm really struggling to think of any game that has surpassed it. The FPS genre has been dominated by titles focused on multiplayer for the last 6-8 years, so AI has been fairly stagnant.


In games, AI is not my area. But, what I've seen is that games try to avoid training-based AI because it is difficult to get predictable, controllable results. Training an neural/genetic/whatever system is great until a designer asks you to code in a few exceptional situations and a few extra guarantees. However, there is apparently some movement into that area going on.

Here's a recent vid about game AI. https://www.youtube.com/watch?v=jse_ZleruJU&t=11m44s It talks about the progression from FSMs, to Behavior Trees, to Utility AI, into Neural Nets and eventually Neural Evolution.


Call of Duty Ghosts if it was the one with bots that acted like real players. I actually thought I was online when my brother handed me the controller because (a) didn't know COD's had bots and (b) they acted like players albeit with far less range of behavior. It was pretty nice experience with me enjoying screwing with them to test the limits of their capabilities.

Half-life's squads were first to have an impact on me like they were realistic. FEAR was first to really put pressure on me without straight-up cheating (or appearing to). It was awesome. I think Ghosts exceeds it, though, in some ways esp realism.


The latest Fallout game has some interesting things in the enemy AI. In particular, enemies will actively try to flush you out of cover by blind firing even if they only have a guess at where you are (keying into the game's stealth system).


F.E.A.R. is one of my favourite games, and the AI was really well done.


Agreed. Great story too, for an FPS. Shame the sequels were rather lackluster.


Totally agree, the game in its entirety is incredibly well made.


HE'S TOO FAST


SQUAD, CHECK IN!




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: