One of the best things about videogames these days is that you can play against your friends, even if they’re not on the same continent as you. With the Forza racing series, Microsoft’s Turn 10 Studios has taken that a step further: Gamers can race against their friends, even when their friends are offline.
Forza Horizon 2, available for Xbox One on September 30, is different from last year’s Forza 5, which aims to offer an exact reproduction of real-world racing. Instead of being limited to tracks, Horizon 2 drivers have access to a huge chunk of southern France and Italy, from Nice to Castelletto (complete with stunning visuals and realistic weather). Even better, unlike the original, Colorado-based Forza Horizon, drivers aren’t limited to roadways. It’s like Grand Theft Auto: You can go nearly anywhere. Forests, hay fields, off-road trails, and pedestrian walkways are all accessible.
The game is great fun, especially for those who don’t have the patience (or skill) for the ultra-realistic simulations of the track-based Forza games. Taking a Ferrari 458 Italia through a field of lavender may be hilariously unrealistic, but that doesn’t make it any less fun, and we suspect siblings will have hours of fun forcing each other into trees at 150 mph.
But not everyone has siblings, or friends who are online and available to play when they are. And racing against a computer opponents programmed to drive a preset course, making every turn perfectly (and identically), is hardly a blast. That’s why Turn 10 created Drivatars, digital players that drive like humans, not automatons. Even better, they drive like specific humans: Every flesh and blood player gets a Drivatar of himself that learns from him and drives like he does. If your big sister isn’t online, you can play the digital version of her instead. (A separate version of Horizon 2, without Drivatars, will also be released on Xbox 360.)
So even if you’re the only human in a race, you’ll have a diverse set of opponents based on your online friends and other players. Some will be aggressive, smashing into other cars and taking every shortcut possible, while others are more likely to jump out of the way if you drive at them, or use extreme caution heading into corners. Sometimes they’ll just make mistakes, as people do, taking a corner too fast and spinning out, or crashing into oncoming traffic. The single-player mode is awfully close to playing with real people, and it’s a delight.
Next-Generation AI
For years, Turn 10 has wanted to use a learning neural network to totally change how racing artificial intelligence works and, potentially, change how videogaming works into the future. The Drivatars use Bayesian learning to develop in the image of their human counterparts: Yours picks up what routes you like to take, whether you drive fast into a corner or brake early, if you bump other cars to get them out of the way or steer clear of opponents.
The goal is to make a computer that drives just like a human. More importantly, the point is to have a variety of driving styles on the track. Dan Greenawalt, creative director at Turn 10 Studios, used the example of an old rivalry in F1 between Michael Schumacher and Juan Pablo Montoya. They had very different driving styles, one very clinical and precise, and the other much looser. This is what the team hopes the Drivatar can create, rather than having a single variety of AI driving around the track in the same way, lap after lap.
Microsoft developed the technology more than a decade ago and put it to limited use in the original Forza game: Players couldn’t see or race each other’s Drivatars. Technology evolved and last year Turn 10 unveiled an entirely new cloud-based Drivatar for Forza 5, proving Drivatars could be much more fun to play against than traditional AI. But the nature of Forza Horizon 2 makes the feature even better.
In Horizon 2, players can drive wherever they want. Stunts like drifting, driving on two wheels, and getting airborne are encouraged. Going off-road is a central focus, while in Forza 5, driving off the track would carry a heavy penalty. So the rules by which the Drivatars are programmed needed to be updated.
“We had to change some definitions for ‘What is a road?’ versus what isn’t,” said Greenawalt. Fortunately, it didn’t take much: “We just turned it on.” Now, Drivatars are imitating their humans, cutting corners and crashing through cafés, anything to get to the finish line fastest. “They can drive through groves and vineyards, drive between trees. We didn’t know what that would be like. We didn’t know what they were going to do.”
Learning How to Drive
The thought that a computer is learning to imitate human behavior may be somewhat terrifying, but in this videogame, it’s sweet. And it’s produced some surprising results. A large part of Forza Horizon 2 involves simply milling about in the game world, without a particular destination. Greenawalt says his team saw Drivatars, without any human interaction, begin acting almost like real people. “You’ll see two Drivatars that will race against each other,” he said. “We didn’t tell it to do that. They learned that by watching players in Horizon do it.”
During the week I spent with the game, I watched a friend’s Drivatar drive around a hay field doing donuts and driving into hay bales. I gave him a heads up that his digital representation was goofing off, and he wasn’t surprised: He’d spent 15 minutes doing exactly that earlier in the day. Another time, I was in second place toward the end of a race and I saw a Drivatar smash into an easy to avoid tree. A traditional AI would never have crashed, but whatever person that Drivatar was based on must have difficulties with arboreal obstacles. It felt like I was driving against real people, who make real mistakes, rather than precise, boring automatons. It was great.
Turn 10 can perform big data analytics on the Drivatars, examining what they’re doing as a population and watching particular behaviors spread. “We didn’t train cars how to block [competing cars from passing them] in studio, they just learned how to block from the population,” Greenawalt said. “It learns the correlation between the behavior and the conditions.”
Unsurprisingly, jerk humans make for jerk Drivatars. Players might turn around and drive the wrong way on a track to mess with their competitors, or smash into cars going around a corner to get an advantage. Drivatars learn to do the same. Turn 10 can change the game code to stop particularly obnoxious behavior, but errs on the laissez-faire side of things. The anything-goes approach is what makes Horizon 2 so fun, Greenawalt said. “We try not to put too many clamps or behavior modifiers on the Drivatars because it takes away from what makes them so unique.”
More Than One Way
It’s easy to see how the Drivatar technology could be applied to other games as well. In a multiplayer first-person shooter like Call of Duty, some players like a run-and-gun strategy, killing anything in their path. Others prefer to sit back and snipe, waiting for the action to come to them. They’re both legitimate ways to play the game, but with very different user behaviors. Having Drivatars (Shootatars?) based on your friends could make gaming much more interesting and could allow gamers to actually trust and work with their computer teammates, rather than simply using them as meat shields.
It could even make gaming more civilized. Human players tend to treat computer controlled opponents like dirt, but in the year since Forza 5 was released, Greenawalt says, the human community as a whole began driving more courteously. Their Drivatars echoed the shift. In online group play, drivers tend to be more polite because there are real people behind the wheels of the other cars. Drivatars have started to close the gap between online group play, where drivers are moderately polite to other humans, and single player mode, where the opponents learn to closely imitate humans. Now, the community as a whole drives much cleaner in single-player, ending up with better racing. “Social engineering intersects with AI and technology,” said Greenawalt.
No comments:
Post a Comment