New fighter: Umbrella! Available on Steam.

A New Perspective on Single Player AI


Written by:

I want you to take a second to look back on your fondest memories of your favorite fighting games. Think back to your most dominating victories, your closest comebacks and your most crushing defeats. Whether you’re a tournament veteran, or just passing the stick around the floor in your living room, all these memories are almost certainly against real life opponents. To even think that beating up on the poor computer could be as exciting beating up on your friends and rivals is just silly. My name is Ian Cox, and I’m a designer on ***Skullgirls***. We’ve recently started working hard on the game’s single player modes and I’ve been in charge of scripting the enemy behaviors. It’s a difficult task, but it’s been a lot of fun, and I wanted to let you all in on a bit of our thought process as we start plugging away at it. In the past, the single player modes have only existed as poor substitute for the “real” game. With the increasing popularity and quality of online play, we had to take a hard look at what the real purpose of single player was in today’s fighting games, specifically with what we wanted to accomplish with our single player AI opponents in ***Skullgirls***. The conventional fighting game AI tries to approximate human actions, which is of course a really hard thing to do in a convincing way. When playing against one another, skilled human players do things like repeat a pattern to trick their opponent in to doing stupid things, look weak to bait out unsafe moves, or train their opponents that they’ll always respond to a certain situation in a certain way, so they can defy conventions at a critical moment. This means an AI that was to be successful at simulating a human would have to be able to recognize your patterns, identify weak-looking positions, or get frustrated. These challenges aren’t impossible, but with a human opponent only ever a few menu options away, we decided to go a different route. Hardcore players often spend the majority of their solo game time in training mode. It seems silly to think that a mode with no opponent at all could be more fun than single player, but in training mode you have the ability to learn all these details about the characters and engine, and learning things is fun. There’s a real sense of accomplishment in learning a new technique or combo. While training mode is great for finding and pushing the limits of a game, however, it’s not so great for teaching you the basics, nor is it particularly interactive, and its blank-slate nature can be paralyzing in its freedom. As we’ve said before, we’re really intent on making sure our game teaches you how to be good at it. So we want ***Skullgirls*** to have AI opponents that teach you some of the things that are traditionally difficult to learn on your own, while capturing the joy that comes along with constantly learning new things and improving your game. To meet our goals of having AI that is both satisfying to beat and instructional, computer opponents in ***Skullgirls*** will use pretty specialized strategies and have stark strengths and weaknesses, especially at lower difficulty levels. They’ll be focused on specific techniques: if we make an AI that is really good at anti-air, you’ll have to learn to beat them on the ground, and if we make an AI that blocks a lot, you’ll have to learn to throw. We can also tune these AIs to challenge or emphasize certain aspects of your character. You can bet that if you play through the game as Cerebella, one of the first opponents will be especially vulnerable to throws, since learning to throw is a really important skill to have as a grappler. You’ll also fight powerful zoning characters, since learning how to get in on runaway opponents is also essential. You might notice that the AI will seem more predictable than you’re used to seeing in other games, because we want to let you know what cues to look for. We want to teach you to keep an eye out for opponents holding a charge, or what kind of follow-ups you should expect after this or that special move. You might think that more predictable AI will mean easier AI. This isn’t exactly true, but it does mean that once you learn the patterns, you’ll probably be able to beat it pretty consistently. Once that happens, you can crank up the difficulty or switch to a new character, and watch the difficulty of the puzzle grow along with your ability to solve it. For example, an easy rushdown AI designed to teach you to block effectively will do simple jump-ins to ground chains to late sweeps. As the difficulty goes up, it will employ more advanced techniques, like using triangle jumps, and low hitting assists. We’ll have to develop a lot of different AIs to both deliver the amount of content you expect, as well as teach the broad swath of skill we want to communicate. This, too, will teach something, as players are going to learn to read the playstyle of the opponent, and change up their strategies to beat them on the fly. One of the upsides to this kind of AI design is that it takes advantage of the way people already play against the computer, which is basically: try a strategy out, use it until it stops working, and then switch it up. In the end, I don’t think most players will even notice the difference between our AI and those that they’ve played against in the past. The only difference is that our AI will encourage you to switch up your playstyle a lot more often, and in a variety of ways. Through this change, we hope that single player fights will stay fresh, and encourage you to learn more about your chosen characters. As we’ve started implementing the first AIs adhering to this philosophy, we’ve seen a lot of success so far. Playing ***Skullgirls*** by yourself should be fun, so we’re designing the opponents to key in to a different part of what makes playing fighting games enjoyable. And while you’re standing over your defeated AI opponent, you can be confident you’re a better player than when you started. Follow [@**skullgirls**](http://twitter.com/#%21/skullgirls) on Twitter, or “Like” us on [Facebook](http://www.facebook.com/skullgirls) for all the latest **Skullgirls** news!