Recently, some players have claimed on the Reddit forum that they have encountered cheaters in Rocket League qualifying matches. In the gaming community, it is not uncommon to accuse others of cheating, but many times it is often a misunderstanding caused by a player's technical gap. However, after watching the video in the post, many people pointed out that this is indeed not an operation that ordinary players can perform, but a "robot".
Evidence video: Youku Video Cloud
The discovered cheater is on the same team as the whistleblower, and the video is the cheater's perspective. Although there are many masters in "Rocket League", his operation is not like those high-end players, who use the air trick to send the ball into the goal, but perfectly avoid the opponent's attack and dribble.
In fact, this form of cheating is true, and it's becoming more common in games: many people have started using machine learning-trained Rocket League bots in their online stacks.
The creators of RLGym, the API that facilitates the training of such bots, say they know their project could be exploited by cheaters, but that is not their intention. In fact, there is currently a Rocket League bot development and competition community that doesn't use the technology to cheat, with another API called RLBot at its core, creating and using custom bots to play against each other in offline games or local tournaments.
RLGym is a game-changer, allowing robot developers to use games as an environment for reinforcement learning algorithms. The core idea is that, instead of manually scripting the robot's behavior, tell the system what it needs: for example, to bring the ball close to an opponent's goal. Rocket League simulations are then run for several hours, during which the robot learns gradually to achieve the desired result.
In fact, the process is not so simple. Aech, a developer at RLGym, said: "Designing rewards is an art in itself. Last week, Aech took a Q&A on the forum about cheating. He said the machine-learning robot Nexto, which is used for cheating in Rocket League, is "very difficult to make," so it's unlikely that there will be a large number of variants. However, there are more and more cheaters, and Aech has previously predicted that someone will train a robot more powerful than Nexto.
Bots themselves aren't new, and according to Aich, the latest news is that someone has used "their own tools to make changes to the game, which aren't as restrictive as RLGym or RLBot are in online games." "That's why Nexto showed up in qualifying.
In an interview with PC Gamer, Aech said, "RLGym is very fortunate to find a super welcoming and inclusive community where we don't let these cheaters stop us. ”
"We're taking steps to keep our robots from being abused in the future, and we can't wait to get into a fight to showcase some of the exciting projects we've been working on."
Currently, if a player and a bot are matched together, there is no other way than to report and do their best. However, it is said that due to some flaws in training, the robot is not very strong when teeing.
Aech concludes: "We've now seen a rather strange phenomenon with several machine learning projects, where robots usually learn how to tee off better early in training, but as they progress over the rest of the game, they all seem to lose the ability to tee off. ”