I was skeptical when I came across a Reddit poster claiming they "for sure" encountered a cheater in ranked Rocket League. Uh huh, just like how everyone who kills me in Rainbow Six Siege is "for sure" aimbotting, right? Then I watched the video. Well friends, I regret to inform you that people are cheating in Rocket League.
The alleged cheater was actually on the same team as ghost_snyped, the Reddit user who posted the clip embedded above, which shows the cheater's perspective for part of a doubles match. I've been playing Rocket League for seven years and I have never seen a human being play like that at any rank. There are masterful Rocket League dribblers out there, but it'd be unusual for a skilled player to stay so rooted to the field—most throw in some aerial maneuvers here and there—and to carry and flick the ball that flawlessly.
Sure enough, this is a real problem: People have started using a machine learning-trained Rocket League bot in online matches.
The creators of RLGym, an API which facilitates the training of Rocket League bots like the one in the clip, knew their project could be used to cheat, but that wasn't their intention. There's actually a cool, non-cheaty Rocket League bot development scene centered around another API called RLBot, which enables the creation and use of custom bots for offline play or to pit against each other in LAN tournaments.
RLGym changed the game by allowing bot developers to use Rocket League as an environment for reinforcement learning algorithms. The basic idea is that, rather than scripting your bot's behavior by hand, you tell the system what outcomes are desirable—the ball moving closer to the opponent's goal, for example—and then run hours and hours of simulated Rocket League during which the bot gradually "learns" to achieve those desired outcomes.
Practically speaking, it's not nearly that simple: "Engineering reward functions is an art all to itself," according to one of RLGym's authors, Aech, who ran a Q&A on Reddit about the cheating problem last week. The specific machine learning bot being used to cheat in Rocket League, Nexto, was created by the RLGym team and was "exceptionally hard to make," Aech said. That means it's unlikely that tons of Nexto variants will start popping up, but now that cheaters have caught on, Aech does expect someone to train an even better bot using the RLGym API.
The bots themselves aren't new, then. What's new is that someone has implemented "their own tools for manipulating [Rocket League] that don't have the same restrictions against playing online that RLGym and RLBot do," Aech says. That's why Nexto is now appearing in ranked matches, something RLGym doesn't condone.
"RLGym is incredibly lucky to have found a super passionate and understanding community, and we won't let these cheaters stop us," Aech told PC Gamer. "We're taking steps to ensure our bots can't be abused in the future and we can't wait to show everyone the exciting projects we've been working on."
In an email, Rocket League developer Psyonix told PC Gamer that it's aware of "a small number of players" using the bot to cheat in Rocket League ranked play. The studio says it's "actively investigating solutions."
It sounds like we're not facing a Nexto epidemic just yet, but I did spot a few other popular Reddit posts complaining about encounters with the bot in ranked matches. The other clips show alleged cheaters playing with the same ultra-exacting ball handling.
For now, there's nothing a player can do if they're matched against a suspected Nexto bot except report the cheater and do their best to defeat the machine. If you need a tip, apparently it isn't very good at faceoffs due to some emergent flaw in the training.
"A fairly curious phenomenon that we've seen repeated by several [machine learning] projects now is that bots will typically learn how to be really good at the kickoff early on in training, but as they improve at the rest of the game they almost always seem to lose that ability to do the kickoff well," wrote Aech.
It really feels like the sort of flaw a sci-fi movie protagonist would discover just before their final showdown with a rogue AI. It may also help players to know that Nexto appears to respect Rocket League's only rule, aka Rule 1. (Although, does that mean we have to accept it as one of our own?)
Late last year, I said that machine learning represents a genuine change in how we interact with and understand computers, and here's another example. As a hobby research project, RLGym and Nexto are super cool. Along with experiments like Google's StarCraft 2 AI, these Rocket League bots predict a future in which games will contain much more capable and potentially lifelike AI opponents (although Nexto's playstyle is rather inhuman). At the same time, we now have to deal with the most advanced videogame cheating method ever known: Bots that theoretically can be trained to master any game, perhaps even mimicking occasional human errors so as to be hard to detect. It's gonna be a wild decade.
This article was updated after publishing to add a comment from Rocket League developer Psyonix.