Cheats and Walkthroughs
Cheats and Walkthroughs
Artificial intelligence in games is a bit of a misnomer. Yes, it's artificial, but no, it's not intelligent. It's still a program, following a decision gate and making choices based on pre-determined outcomes and priorities. But when these become complex enough, they can appear to make something sentient. That's something important in every video game where you're playing against the machine, which is pretty much every game if you think about it. Whether you're facing enemies in an FPS, or fighting off monsters in an MMO, you'll want smart foes to compete against, or there's no challenge to the game.
This is where programmers step in and do their work: building systems, paths, traits, and databases that will inform enemy constructs about what to do in nearly every possible situation in the game. This is why the AI Game Programmers Guild produces an AI track at the Game Developers Conference in San Francisco, bringing wisdom from those who have gone before to the up and comers. At this year's GDC, those three wise man were Michael Dawe from Big Huge Games/ 38 studios and his work on Kingdoms of Amalur: Reckoning, Daniel Brewers from Digital Extremes on Darkness 2, and Borut Pfeifer from Plush Apocalypse Productions with Skulls of the Shogun.
First of all, let me tell you that building AI is hard. You don’t just simply buy the AI add-on pack for your game and drop it into the code. Although it you are out there tinkering with AI and game development, Michael Dawe recommends using optimal reciprocal collision avoidance (ORCA), which actually is a plug-and-play formula for collision avoidance devised at the University of North Carolina. After all, you don’t want your baddies constantly running into each other.
First up was Pfeifer who explained the many lessons he learned while working on the AI for Skulls, and there were many because he had never worked on AI before. He also wanted to point out the their game has been delayed until the fall, but not due to AI. Instead, it’s because of delays caused by the platforms, and the fact that they are adding some multiplayer elements to the game.
Skulls of the Shogun is an RTS game with board game elements, so Pfeifer had unique challenges like figuring out how to get the units to decide between combat or going for resources. He says “It’s not about building a grid or the perfect square. It’s about tactics, and less about rock, paper, scissors.” Meaning that the units need to learn what the optimal responses are in situations, and not simply choosing from an “If that, then this” approach.
This slide sums up what he learned from working on the AI, and it also gives you a very surface level hint about how tough this part of game development is.
Daniel Brewer from Digital Extremes was up next to talk about Darkness 2, and he called his portion “Building Better Baddies.” While much of what he had to say was about collision avoidance and the perception system they used (a cone-based line of sight system), one of the more interesting things they did was to establish what they call a “bark” system. This is where an enemy agent might spot you on the edge of their vision, and then call out, or “bark” a response so that the player knows what the AI is doing. They even worked it out so that nearby agents won’t be talking or “barking” over each other during events.
He also explained that they have a reaction time ramp, so that there is a gradual change between an enemy becoming suspicious, and then heading to check out a noise or something he might have seen. “It’s so you don’t just peek your head around a corner and instantly everyone starts charging for you.” It’s more forgiving to the player, and results in NPCs not having instantaneous state changes.
Interesting tidbit: the Darkness didn’t seem to interfere with the AI that much, except for having them spot the Darkling from time to time.
Ending the session was Michael Dawe from Big Huge Games / 38 Studios to talk about Reckoning, and an interesting (and perhaps somewhat forgotten) fact he pointed out was that Big Huge’s last console game was Catan for Xbox Live back in 2007. Thankfully, Kingdoms of Amalur proves that they know what they’re doing. Actually, they have to since the game is comprised of over 600 spaces, 24,000 different objects, and over 500 quests. That’s a lot to deal with before you even toss AI into the mix.
He also pointed out that Reckoning is not Fable. They don’t have the NPCs keep track of things or complicate things with an honor system. Which would also have affected the way NPCs act in the game. His challenges were more about scheduling the NPCs, working on the designer / programmer interface, and scaling the difficulty in the game in how it related to the NPCs.
They were also developing the engine for Reckoning at the same time they were designing the game, which led to additional challenges for the team. They originally built huge maps of the connected spaces with instances for everything an NPC could interact with, but it ended up being a much bigger job than they expected. They ended up using what Dawe calls a Belgian Waffle Grid to determine behaviors, based on the “weight” assigned to them.
He also wants to take AI “beyond the Kung Fu circle,” referencing how in old Bruce Lee films he would be surrounded by a gang of baddies, but they would only attack him one at a time. They’ve modified NPC behavior in Reckconing to allow for multiple attacks, depending on the NPC priorty.
He included a piece of what I can only guess is programmer behavior that got a huge laugh, mentioning that given freedom, designers will do everything. His example was “Why is wander_behavior causing a 20ms spike?” This brought a huge laugh to the room, and sailed completely over my head. Proving that while I know good AI when I play against it, I’d probably never be able to design it. But maybe you will! Check out the Game AI website and see what you think.