FIFA 22’s HyperMotion is ‘the beginning of machine learning taking over animation’

As Sam Rivera explained it to me, the success of FIFA 22’s new animation technology will be seen in what wasn’t recorded during a groundbreaking motion-capture session — involving 22 players all playing a start-to-finish game of soccer — earlier this year.

“We started working on an algorithm about three years ago,” explained Rivera, FIFA 22’s lead gameplay producer at EA Vancouver. “What that algorithm is doing is learning from all the data for that motion capture shoot — how the players approach the ball, how many steps do they do to get to the ball, is it three long steps and one short step; what is the proper angle, with the proper cadence, to properly hit that ball?”

Then, Rivera says, “it creates that solution, it creates the animation in real time. That is very, very cutting-edge technology. This is basically the beginning of machine learning taking over animation.”

In a dozen years of covering sports video games, I’ve found their near-annual claim of new animations, often a record-breaking, more-than-ever number of them, to be a marketing cliché. Recording a bunch of snazzy acrobatics doesn’t necessarily mean regular players will see any of it. Almost a decade ago, NBA Live designers told me they sifted through the remains of the abandoned NBA Elite 11 and discovered tons of amazing blocks and dunks that the game’s engine would simply never serve up. The lesson: Recording this stuff is fine, delivering it is where the rubber meets the road.

In FIFA 22’s case, their motion capture team might have caught the authentic anguish of a goalkeeper beaten by the deciding score, in a 4-2 match between two real-life rivals from Spain’s Primera División RFEF (the nation’s third tier). But neither that, nor CD Gerena’s desperate play leading into Atlético Sanluqueño’s crushing counterattack, should be the reason players at home feel they’re watching a more realistic presentation of soccer, or finding it more fluid and responsive as they play it.

HyperMotion, as it is called — because, yes, marketers gonna market these animations, too — allows Rivera and his development colleagues a have-it-both-ways proposition when it comes to player movement and interaction, he said. “In the past, we used to prioritize short animations, so the game is responsive,” he explained. “If you are in a long animation, the game looks good, right? But if the situation changes [in the middle of the animation], there’s a defender coming, you’re stuck, you’re probably going to get tackled, and that doesn’t feel good.

“With access to HyperMotion, we’re able to put in there longer animations, longer ball control animations,” Rivera said, “but the technology allows us to transition in the middle of the animation to a different type of animation, if the situation changes.”

More concretely, HyperMotion is going to be seen in things like the defensive formation maintaining its shape and moving in concert more consistently. That will likely be the first and most visible evidence players will see of the all-22 recording session in the new game. In previous editions of FIFA, the AI was typically directing players in groups of two, maybe three, depending on who was closest or whether someone was manually called in for support.

Players of FIFA 22 should find the defense tougher to pick apart or challenge one-on-one. And in response, players off the ball on offense will move more naturally to support a run, perhaps ameliorating a common frustration felt in higher difficulties or online multiplayer.

“Now that we’ve seen some feedback on FIFA 22 and the closed beta, people are definitely calling out that there’s a lot of features that are making the game different — it feels different,” Rivera said. “We were recording something with our pro [esports] players, and they were trying to play a game similar to FIFA 21. And they were not necessarily scoring many goals. Even the chance-creation game, you need to think a little bit different. All the things that the animations changed, that we changed in the game, in terms of positioning, very different experience. And that [experience] was, ‘It feels fresh.’”

As Rivera said, this is “the beginning of machine learning taking over animation,” and EA Vancouver’s FIFA team is aware that their work could have applications in other EA Sports products, if not other Electronic Arts games at large. In the first year of implementing HyperMotion, though, Rivera said his team is focused entirely on getting this right within FIFA 22, and less about using that as some proof of concept.

Most of Rivera’s work implementing HyperMotion wasn’t in the algorithm (that was done by an advanced-works division at EA Vancouver) or the motion capture. Rivera and his team were working on building pipelines and processes to accept, interpret, and use whatever their machine learned and spit out. Those processes are specific to their game, he said. If Madden NFL designers attempted the same thing in their sport, they’d have to develop their own. And Rivera’s first customer is the FIFA 22 player, anyway, not another developer in his company.

“At a high level, what we heard the most is that people wanted more differentiation” among the players on the virtual pitch. Every year has certain quibbles, problems, and exploits, he said, but “we knew what people wanted was a jump in terms of differentiation and realism, making sure the game looks better. So we knew that we needed a bigger investment in terms of animation and technology.”

That took three years, capped by 90 minutes on a field in Spain. “You multiply that 22 times, that’s what allows us to bring to the game over, what, 4,000 animations just in FIFA 22?”

Ah, there it is! The animation count. But Rivera is legitimately proud of it, all the same. “It’s record-breaking for us,” he says.

Roster File is Polygon’s news and opinion column on the intersection of sports and video games.