Sharing the Sandbox: Can We Improve on GTA’s Playmates?

Dave Mark on May 6, 2008

For this week’s developer discussion on, Dave Mark looks into the recent best-seller Grand Theft Auto 4. Let him know how you think the AI could be improved by posting a comment below!

Let’s face it. There is only one game on the radar these past few weeks. The release of the latest entertainment excursion into criminal mischief has gamers and non-gamers alike a-buzz with the usual dichotomous din of praise and vilification. From the seat of a game industry professional, releases such as these are interesting in a different sort of way. As the avalanche of reviews, both paid and amateur, come rolling in, game developers of all stripes start digging through the fluff and the furor. We are always searching for the sometimes obscure fibers of opinion that almost take on the role of forensic evidence. And to what end are we perusing this cavalcade of critique? To find a clue that may lead us closer to the holy grail that is “customer satisfaction” in our industry — what do people either worship or abhor about the latest and greatest title on the shelves? Put simply… what do we need to focus on?

The Good, The Bad & The Unscripted

Certainly GTA has set a new standard in a lot of ways. Visually, it is stunning. The world is immense despite being somewhat smaller geographically than San Andreas. Many of the comments talk about the excellent physics engine and the vehicle handling and dynamics. The content is seemingly endless. The potential dialog possibilities are enough to have been made mention of by several reviewers. At least one even went so far as to say that riding around in cars with sidekicks actually makes for entertaining and informative conversation. When was the last time that people actually enjoyed killing time in what is ostensibly an action game just to listen to the dialog?

Grand Theft Auto 4 AI

Screenshot 1: Semi-crowded docks in GTA 4’s Liberty City.

There are the detractors, however. While some people parrot the propaganda about the new cover system (both for your character and the AI agents), there have been plenty of anecdotes about characters selecting the wrong side of a corner to hide “behind”, leaving them mercilessly exposed. Another I saw in a GTA-related thread here on the forums (registration required) told of some cops who, upon having the door of their cruiser blown off — behind which they had been taking cover — actually run around the exposed front of their car in order to open the other door and take cover once again. While I applaud the notion of the agents actually creating a new cover point, the fact that they needlessly exposed themselves in order to get there bursts the tenuous bubble that is the coveted “suspension of disbelief.”

There are some spectacular (and somewhat amusing) errors as well. I was treated to a YouTube video wherein the player was hiding on a fire escape. A seemingly endless stream of officers are in such a hurry to follow the player to his perch that they hurtle themselves off the side of the building, over the fire escape, and to the pavement below. A technical triage would probably determine that the nav mesh lacked a large warning sign saying “you probably don’t want to run between these 2 nodes”. Therefore, the policeman agents’ running momentum launched them beyond the node on the fire escape. Had they been walking, they would have likely stepped down. (Note, I can’t verify the legitimacy of this video, nor can I offer more than just a guess as to the cause. It’s hillarious nonetheless.)

Another complaint is that, while the bystanders are engaging and even interesting to follow around and watch (if you can resist beating the snot out of them that long), there aren’t near enough of them to make for a convincing urban area. Where are the massive throngs of people on the daytime street corners? If you drive up the street and up onto the sidewalk (pavement for you Brits), shouldn’t that fabulous physics engine be given the chance to shine via the resultant human bowling alley? Instead, the inhabitants are a bit sparse for a thriving metropolis — another slow leak in the believability bubble.

Not ‘What is Missing?’ but ‘Why‘ is the AI Missing?

I want to redirect myself now lest I digress into a review of GTA IV (especially since I haven’t actually played it yet… but when has that stopped even a professional reviewer?). My point in this article is somewhat more muted and hypothetical. Sandbox games — or at least free-roaming RPGs — are becoming more and more prevalent of late. With the likes of the GTA series, Assassin’s Creed, the Fables, or Saint’s Row, the latest cool thing to do is develop a massive open world where the plot is almost reduced to a mild suggestion. But, there are reccurent themes of developmental difficulty in those projects. They are sometimes addressed and even more rarely solved with varying degrees of success. One thing that is not lacking, however, is the overflowing suggestion box from the gaming public. A sampling of those polite requests indignant demands would show that they often begin “How come you can’t do ——- like they did in ——-?” or “Your game would have been better if you had included ——-” Do these people really think we developers are smacking our foreheads and saying “Wow! We didn’t even think of including that [really obvious feature] in our game!

Grand Theft Auto AI

Screenshot 2: Pedestrians interacting on a sidewalk in GTA 4.

As much as we don’t like to listen to the largely uninformed ramblings of the masses of gaming peasants, there are some points that shouldn’t necessarily be dismissed. In fact, I can see how there would be some confusion and even wistful longing from our clientele as they look at features in one game and wish that they were in another. For example, people enjoy the crowds in games like Assassin’s Creed and Dead Rising. In the former, the crowd is an integral part of the environment. In the later, the crowd is an integral “character” in the game. And yet, as I touched on above, GTA 4 portrays what is purportedly an image of New York City and makes it look like the daytime “ghost-city” version from the movie “I am Legend” by comparison. It would be natural for the gaming public to want to see similar throngs of people milling about in a GTA-style game. Some of the AI middleware solutions and industry whitepaper demos are pushing thousands of simultaneous autonomous agents around. Is it reasonable to ask for that technology to show up in big games?


Screenshot 3: AI traffic avoiding AI characters in Liberty City.

Other AI failings are more subtle. In the industry, we all pour over the “AI Game Programming Wisdom” books and the latest articles here on AIGameDev addressing issues like determining cover points, moving through threat areas, dynamic pathfinding, and squad tactics. Some of the techniques that we exchange are becoming more prevalent, some are still a bit more esoteric. The point is, we are slowly conquering the ills that ail our AI. And yet, one place where they seem to be less likely to show up is in our massive marquee titles — such as a GTA 4. Why is that?

One theory could be that the big titles are big specifically because they are focusing on the visual and visceral “splash” (which sometimes includes visible splashing viscera). In the past, the complaint was that games were spending so much time on the graphics and story-driven content that the AI became almost an afterthought. The claim was partially that games that did this became the big sellers because of that focus. The other side of that chicken/egg argument would be that the developers took that approach because they thought it was the formula for a successful AAA title. Either way, the game reviews belie that premise. Gamers do notice the weak AI… and they are becoming more and more vociferous about it.

The $100-million Question

Grand Theft Auto Sandbox

So, as usual, I ask the question… what is holding us back? We can only cite limited technological resources for so long. The hardware is almost expanding faster than we can utilize it. (I’m being a little facetious there — save your emails.) Development budgets and ship dates are, as usual, a constraint. However, with people shelling out well over $60 for AAA titles in some cases, the market’s willingness to fund our enterprises may be less of an issue than we fear. Rockstar allegedly spent $100-million on GTA 4. Methinks they will make that back.

Perhaps, from a non-business standpoint… that of simply an AI developer, we should be asking ourselves what the challenges are in bringing all the top AI techniques together into the massive game environments that are so en vogue. What is the bottleneck? Is it money? Time? Hardware? Technology? Unwillingness? Unimaginativeness? A belief that those features are not wanted by the gamer? Or is it simply fear on the part of AI programmers to undertake those steps necessary to put that much life into such a massive world?

In the end, do we need to worry that the sandbox is getting a little too big… and a little too lonely? Post a comment below.

Discussion 4 Comments

Jare on May 8th, 2008

Ambient characters in GTA4 are more detailed and more varied than in previous GTAs. They are also more numerous. Let's say 2x each for a total of 8x. Seems to me a pretty good improvement over the previous generation.

zoombapup on May 8th, 2008

Perhaps one of the things thats holding AI development for games is the lack of a proper means of dissemination and sharing within the industry. What I mean is, is that we get lots of reasonably vague articles, books devoted to a particular subject but without sound "game like" implementations. Essentially we arent sharing the techniques that work with each other, so essentially everyone is working in isolation to solve the same problems, rather than together to forward the field as a whole. Of course there are always exceptions, like the FEAR SDK and such, but in general, I'm not convinced that many people are actually developing much that is unique, they are basically just implementing the same things. Maybe that is a commercial pressure in itself (AI being intrinsically linked to gameplay after all).

Dave Mark on May 8th, 2008

[quote=zoombapup;2420]Perhaps one of the things thats holding AI development for games is the lack of a proper means of dissemination and sharing within the industry.[/quote] Hey Phil, keep this under your hat... but there is major movements that is going to try to address that very issue. I can't say more since I'm not the one running the show. Ack! Did I say that out loud!?

diegix on May 8th, 2008

When considering the AI, or any other game feature, I think we should consider their domain first. We can argue that since Gears of War have cool cover system (though I personally saw the AI characters cover in the wrong side of the cover point...) everygame that come after it should have the same. Since Dead Rising have 100 zombies, all games must have 100 characters or we are held back. That's almost like saying that a sitcom should have the special effects of the last big action movie. From my perspective when you are making a game you have to choose where to put effort into. Following with the example of the ghost city, you can have 20 pedestrians with high level of detail and behaviour or have 100 with low level of detail and then you choose one. GTA chose to improve in number and detail instead of put all the eggs in number of pedestrians. Same thing can be said about other features, if you make an open game like GTA you cannot expect to have complex behaviours/situations like in Call of Duty 4, because they are completely different domains with different complexities.

If you'd like to add a comment or question on this page, simply log-in to the site. You can create an account from the sign-up page if necessary... It takes less than a minute!