So this bot actually is beatable, or at least hittable, if it works the way he claims it does. He claimed that the bot will attempt to grab you if you are shielding. But, since some characters in this game have OOS options that are faster than 7 frames, if the bot tries to grab you when you are playing one of those characters, and you then do that OOS option after he has started grabbing (in which case he's commited and can't do something else), your OOS hitbox will beat his grab, thus preventing a JV5.
gamerfreak5665 yeah fr any update? So many people in all of these videos’ comments sections saying they want to run smashbox against TAS. Was it successful? Is the video anywhere on RU-vid??
The reason why fox bot can be so good is because fox simply just has options which are simply not punishable like shine. Low tiers wouldn't really be able to be programmed in this way since they would have counterplay and since the bot is predictable they wouldn't be that good. Unless they were training with learning
@@shayboual1892 A lot of moves aren't punishable if a computer is playing lol and even shine would be punishable if a computer was on the receiving end of it.
Smashbots fighting smashbots is impossible. It only works on player 2 and assumes you're player 1, and it against tas would be smashbot against a perfect player.
in fact it can play against math/fox/falco/samus (for now) he knows those match up but it has fram data of every char, the only thing that counters him is counterpick stages, it can only fight on sd) Go watch the full video of smashbot vs pro, it's in vgbootcamps vod. so a tas could be made against tasbot.
Hi. First of all, i would like to say thank you for brining AI to Smash. i have one concern tho, and i'm sure you've heard it many time: there should be a human-like reaction time implemented in a bot (or a new/different version of it, that would start to learn all over again with the reaction time). Main issue is, when we think to AI in games, reaction time isnt an issue that has been talked a lot about. In Chess and in Go, the game is turn by turn. Therefore one can take as much time as he needs to (i know it's within 2 hours limit, but it is still long enough) in order to play. Then the bot can play, as fast or as slow as he wants to. The fact is that the game is not reaction-time based (of course there is a 2hours limit) but it is turn by turn based, which makes reaction time almost irrelevant. In the game of smash. And almost every video games that isnt turn-by-turn based, reaction time IS PART OF THE GAME. If you take it off, it is not the same game at all. The strategy of the game, the meta game, how you move, when you decide to attack, etc. every timing is based on your reaction time. Reaction time IS PART OF THE GAME. This is why reaction time should be implemented in order to make a really interesting bot that we can learn from if it is even possible with average computational power, to make an AI that can beat the pros. (Because even if it is cool to watch a frame-perfect infinite waveshine, we all know it is not realistic and it gets boring to watch/play against after a while) A reaction time of about 0.25-0.3sec should be implemented. Even this is quite fast since it will have almost instant option selection and action (so i think going to 0.3sec should be the best). (Note that the reaction time only apply in reaction to your opponent action, not to yours (for exemple, if ou are inputing an input sequence, it could be frame perfect between each of those input, since you already knew you were going to input those)) Hope you've read this, and hope to see a bot like that soon! (also with videos showing his evolution would be crazy interesting)
pcrsweetness Exactly. This guy put a few months into designing smashbot. A learning algorithm is outside the scope of his skills and would take a team of people to implement. When AI tools become more available and computer processing becomes cheaper, we can consider building a learning smashbot with human reaction time. Right now anybody with those skills to create that is going to be using them on a real job making good money. We will have to wait some years before seeing a smashbot that learns and performs akin to OpenAI DotA project or what deepmind is doing with StarCraft. Both those projects are composed of entire teams of people who are more or less geniuses when it comes to AI systems. That's what it takes to create an AI a step up from this, one that could beat top players with human reaction time constraints factored in. It's an entirely different order of magnitude in difficulty to create what OP is proposing.
I dont know irf it would matter, but I dont know why they didnt get any really good pro to play it but all well, or is this just showing the jv5s and there were other sets that went better as well?
It doesn't matter. The bot has an instant reaction time and perfect frame data. If you throw out an attack it will simply power shield of spot dodge into shine The only way to hit it would be to frame trap it, which isn't possible in neutral against a shine only Fox. It doesn't matter who fights it, they literally can't hit it The only thing they can do is pick Fox/Falco and clang shines best case scenario
Personally i would have rather seen a smashbot with 15 frame reaction time implemented, because it seems kind of boring to have a bot that just reacts frame 1 to everything
Who are these "pros"? And why the heck do most of them not even seem to acknowledge what they're doing? You're fighting a god AI. Have some fun. It's not serious. You're not supposed to win.
Chillava Lmao they're top level players and well known players in the community, and they seem like they are having a lot of fun. Of course they want to see if they can beat or exploit it, they're basically helpless against this bot and they know it, I saw a lot of silly fun stuff in the video. I don't know what you expected from this video, but I saw a lot of fun and personality in this video.
@@pwnography577 Not really, the game froze every time the players were winning, so this bot is beatable when not freezing 24/7 the full 2 hour clip shows how vulnerable the bot really is.
