Not sure what the connection is between YOLOv4 and difficulty of detection? Isn't YOLOv4 about object detection? No matter how good your object detection is, you need to read the game and influence the input which is the part an anticheat is trying to detect, no?
There are probably plenty of ways to tell "fake controller input from a bot" from "actual human controller input" today.
Sounds like a scary arms-race, though. At some point the bots will probably be very hard to distinguish from "skilled human."
It will be sad if online play becomes only enjoyable with people you already know. I'm surprised how gleeful some people seem to be about this sort of "soon cheating will be undectable!" tech advancement.
Or you just go all-in on the surveillance path and there are models that look at your performance in the game over time, your performance in other games over time, etc. Mediocre player suddenly amazing? Probably a cheater! etc... Not great sounding privacy-wise, but Steam probably has access to the data to do this.
There's a form of cheating I've heard about a while back called softaim. Basically the cheating software doesn't aim for you, but it can tell if you're aiming at the person and pull the trigger for you.
The YOLO stuff combined with softaim is going to be pretty hard to detect. The game can't tell if your video is going into the cheating device. Even if it can tell if there's a secondary input coming in for the trigger... people could just mod their mouse to take external input for the button. Someone pathetic enough to cheat absolutely would do this.
I honestly don't know how multiplayer will even work in a year's time or maybe even less!
> I honestly don't know how multiplayer will even work in a year's time or maybe even less!
It's easy, IMO: remove the incentives for cheating. If this is the only way forward, I might be more likely to actually participate in the industry, because it'll put the focus back on intrinsically-fun games, instead of treating games as merely a vehicle for chasing status/rankings/items/etc.
Multiplayer FPS (PvP) games are my favorite - could you explain how they can remove the incentives therein? Almost the entire incentive behind PvP games is beating the other person in any given battle/arena/skirmish. Even if ranks/bonuses/items were detached from "skill", most of the incentive still remains for PvP. Having encountered a number of cheaters over the years, I know they still get joy out of winning the objective of the game with no obvious side benefit.
I used to love playing multiplayer FPS games, without worrying about collecting rare items or my global ranking, because they were fun (e.g. Quake 3). Yeah there's still a "local" ranking (within a game), but the incentive to cheat is a lot lower since it's so localized. Yes people will still cheat, but some of these insane cheating methods won't be worth the effort.
I stopped playing most modern games because they stopped being intrinsically fun. I'd like to enjoy them again.
Some people will cheat in any multiplayer game, even when there are no persistent "rewards" for winning. They are just maladjusted losers who get off on being the center of attention at everybody else's expense.
Competitiveness is never going away its as human as breathing.
The way to defeat cheating is giving users exaustive options to watch and monitor other players and report them effectively. Half the damn games dont even have this sorted out.
You have a report system that weights users honesty based on usefulness of previous reports - so people.who just report good players get downweighted and their reports count for less, then its just a statistics exercise. Combine this with easily identifiable data for things like headshot % to help highlight players for closer review.
I don't have insider info and it's been a long time since I've played, so I apologize if my info is wrong, but my understanding is Riot Games tried to do this with League of Legends and ended up implementing a kernel mode anti-cheat system instead. Presumably, the reporting system either wasn't very effective or it was too expensive to run.
Is that really a way forward for the people who play games most affected by cheating? I can’t imagine CS:GO players flocking to Animal Crossing just because there aren’t aimbots in that game.
I get there are a lot of people who care about that, and that "remove the incentives" might be unpalatable for them. But it's definitely desirable for me, because I don't enjoy chasing social statuses in a gaming universe.
I bet at some point in the next 20 years we will be going back to game rooms, so you go to a place where you pay to sit in a fixed PC with no available USB ports or any way to use cheats, including cameras to catch anyone whose hands movements don't match his digital input, and there you play against other people in the same network/brand of game rooms (not necessarily the same physical location).
Just some of us, meanwhile the wealthy ones will live in semi-closed environments free of most damage made by climate change and other environmental issues.
> There are probably plenty of ways to tell "fake controller input from a bot" from "actual human controller input" today.
Only in the "do these movement patterns appear human-like or not" sense. These aim bots can use assistive devices to input mouse movements and there's no way to tell whether or not there's a human hand moving the mouse.
There might be for now. You just have to train a second model that actually moves the cursor towards the target on normal player behaviour, eventually it becomes essentially perfect, and what then?
As far as performance improvements, oh it's going to make the cheat programmers even more money as they implement a skill ramp up period and get to charge more.
The only way in which I'm "gleeful" is that it might finally put an end to the spyware when we realize that controlling someone else's computer is a losing proposition. Otherwise yeah it does suck.
> "It will be sad if online play becomes only enjoyable with people you already know."
Already long since passed that point for me quite a couple few years back.
Thankfully, it's not that hard to build up a little circle of gamer friends who are fun to play with these days. The messaging options alone currently available are many and featureful, making it dead easy to gather a little group and keep in contact with them to organize a game session any ol' time.
> It will be sad if online play becomes only enjoyable with people you already know.
For me this has always been the case. I've felt the same since the days of Kali. There's enough jagoffs running around in games to spoil playing with the normal people. To the jagoffs, they're playing the metagame of cheats/trolling and the game itself is incidental.
Also, there is enough variance amongst human that eventually the generator network can actually slip under the noise floor and become 100% undetectable.
How about: every time you snipe someone from an implausible distance, you have to identify some crosswalks and traffic lights before the score is counted.
That's kind of beside the point. Classic aimbots work by reading the game's memory, which means that you can aim at things that are on the other side of walls, or behind you, or obscured by smoke or darkness, or so far away that they don't even render. "Only" being able to aim at things that are actually visible is a significant step down from what we have today, which is the tradeoff for being almost impossible to detect.
Aimbots only aim at things you can see, it's one of their main features. That way you won't snap into someone's head through a wall which would make everyone know you're hacking and get you banned.
I'm no expert here but surely aimhacking originally implied wallhacking? The alternative requires doing actual image analysis in realtime, which AFAIK has only become feasible in the last decade and is way more processor-intensive than just reading coordinates in memory and doing some trig to adjust your aim height. Maybe this is selection bias talking, but I see plenty of videos of cheaters who are obviously just snapping to heads through walls.
You don't actually need image analysis. There are tricks you can use to decide if an enemy is visible or not without it - obviously the game itself does it already to decide if it should draw the enemies to begin with :)
Also yes these were invented a long time ago because people got caught snapping through the wall when CS servers started recording demos.
Since you can check in memory where the enemies head is, you can obviously also check if theres a wall inbetween. Cheats for video games have had humanizing/cloaking forever.
That's not true. They don't send data in _regions_ your client can't see/hear, but if an enemy is for example close to a corner or on the other side of a door, you can have a wallhack see that as the client is still receiving it.
Thanks to latency & the realities of the internet, the server has to bias towards sending data the client can't strictly see/hear at that exact moment, but potentially _could_ see if they move. Which is enough to give a wallhacker a meaningful advantage still.
But what you can't do anymore is see where all the enemies are going at the start of the round and run to the weaker part of the map or whatever. So reduced effectiveness, but still a thing.
Only if it’s running on a separate computer and you’re passing your own mouse movements through that computer. I think it’s still typical for the ML aimbots to run on the same computer?
There was a recent demo of one that didn't. With modern hardware you can run Yolo on an RPi 4 with an 8$ capture card and a Teensy as USB HID for like 50$, you could definitely charge 200$ for it as a cheating appliance.
> "BRB I'm gonna ask for VC funding (just joking)"
I dunno … Mebbe you should run do that right now, before someone else does … Get that price down by around half and most every gamer everywhere can afford your shiny new "modern game genie"; Once everyone can cheat, the original cheaters have nowhere left to hide. They're on even footing with everyone else.
