Gamasutra is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Gamasutra: The Art & Business of Making Gamesspacer
Intelligent Brawling
View All     RSS
October 27, 2021
arrowPress Releases
October 27, 2021
Games Press
View All     RSS
If you enjoy reading this site, you might also want to check out these UBM Tech sites:


Intelligent Brawling

February 19, 2009 Article Start Page 1 of 6 Next

[How do you make a great third-person brawler? THQ's Smith cross-examines titles from God Of War to Ninja Gaiden and beyond to analyze the hot genre, in an article originally published in Game Developer magazine late last year.]

Often the hardest part of game development is taking the time to pause, reflect, and determine what is and isn't working. One of the best tools for this analysis is looking at what other similar games are doing.

Too many games are developed with blinders on, and the developers only look briefly and casually at previous titles in the genre. There are countless lessons sitting out there, waiting to be learned.

I'm a creative manager at THQ, so it's my job to help external developers make their games more fun. CMs work with project managers who make sure things get done on time and manage everything else-we creative managers are pure quality control. I recently worked on the next-gen third-person swords-and-sorcery Conan game with the talented folks at Nihilistic.

Let me take you back to about a year before launch. Development has been going well: The controls are fun and easy to pick up, there are plans for lots of unique content, there's a first playable that people enjoy, and everyone's excited about the promise of this title.

But around the time of the first playable (aka vertical slice, aka razor edge), I'm becoming more and more aware that our AI is not really there yet. This is no big surprise, since it's early, AI is hard, and these things take time, but I wanted to give Nihilistic some clear, detailed feedback on what they could be doing to make it better.

I wanted to be very certain that the changes I suggested would be actual practical needs, not just theoretically good things. For example, I'd been asking for good gating - different ways to keep the player from running past enemies without fighting them.

But the more I thought about it, the more I realized that many games do fine without good gating. I flew through the last half of Halo 2 without firing more than a few shots. I wanted to find out which things really matter. And luckily, I had a little bit of free time, so I went to the source to do a little research.


I picked out some of the top games in our genre, and loaded them up. Title selection was somewhat erratic, based on what I had lying around, but I made sure to include a few well-reviewed commercial failures like Mark of Kri. Those games often have the best ideas to steal. Most - but not all - were games I'd played previously.

A few games that I tried were too far out of my genre to apply. For example, the fights in Spartan: Total Warrior were just too big, with dozens on each side. Heavenly Sword came out after the analysis for Conan, but was added after the fact.

In each case, I would get to the first significant combat encounter that had at least three enemies active at once. I would play this encounter repeatedly, spending most of my time doing nothing but holding down block so I could just watch the enemies and how they acted.

For a few of the specific questions, I needed to move around or attack or otherwise break from block. And with some games, I would go a few more encounters in, especially if I knew major new combat concepts were added fairly soon.

Yes, this is not how most players play. But I wanted to distill the AI down to its simplest, most repeatable state. To balance the block-focused bias, I made sure to also spend some time using normal blocking and attack strategies to see if the AI changed radically.

In a game without block, I'd suggest finding another repeatable strategy that lets the AI (or whatever is the subject of testing) do its thing without the interference of constantly being damaged or killed.

I came into the tests with a list of questions based on my previous experience and my concerns on my current title. As I played, I developed a few new questions and had to go back to previously played games to see how they handled those situations. All process is iterative.

Attack Groups

Questions: How many enemies fight the player at once? How do they organize themselves around the player?

God of War (PS2): Enemies fall into clearly two separate groupings-a close group within a weapon's reach of the player character, and a far group a few meters away. All attacks come from the close group. With zombies and harpies, the close group is limited to three enemies at a time. I sometimes can get four near, but it corrects itself quickly in that case. The far group contains 12-15 enemies.

Enemies cycle from close to far fairly regularly, possibly on a timer, possibly just as a side effect of natural movement-hard to tell. Enemies in the far group pretty much just stand there, only switching to the close group if a gap appears.

Mark of Kri (PS2): Two groups, a near group and a far group similar to God of War, but the near group is only one enemy at a time. The far group contains about three enemies in these early encounters.

The near enemy stays near for a while (up to a minute) and makes multiple attacks with time gaps between. Then he goes back to the far group and someone else takes his place. This change of guard normally feels organic, because players tend to move and thus approach enemies in the far rank, giving that enemy an opportunity to switch. But even without movement, the change still happens, just in a less organic manner.

Occasionally, one enemy from the far group makes a charging attack even while someone else is in the near group-this is just a single attack, after which he returns to the far group. This charge can also be used as a way to enter the near group when the near group is empty, settling into the normal pattern after the charge ends.

Genji (PS2): There's just one far mass of enemies, relatively far from the player. Not too many enemies in this group (3-5), at least at the start of the game. One enemy chooses to approach from that group, and then he walks slowly towards the player character. Once he is near the player, he makes one attack (multiple attacks for special enemies), then walks back to the far group.

Prince of Persia: The Two Thrones (Xbox): There is just one group of enemies fairly near the player. One enemy chooses to attack, makes a single attack or a combo, and then goes back to the group. Enemies usually come in fairly small groups; about three enemies at a time.

Ninja Gaiden (Xbox): There is one group of enemies, usually three or fewer at a time. One enemy chooses to attack, makes a single attack or a combo, and then goes back to the group.

Heavenly Sword (PS3): There are more enemies than most games-up to 20 at a time-but still grouped in a near set and a far set like many of the other games. The near group is at most four enemies, and enemies in the far groups can still attack.

The positioning, especially in the far group, is widely varied, with enemies appearing more like a jumbled mass than a clean circle. This variety feels good and makes the grouping less obvious without confusing the player, since the near group is usually clearly defined.

A screenshot is taken from the old build of Conan, showing the inner and outer rings where the AI position themselves. In later builds, enemy position on each ring was more varied.

Conclusions: The solutions used here are more varied than I expected. All the games group their enemies into one or two groups around the player character, but how the enemies attack from their groups varies a lot from game to game.

Prince of Persia and Ninja Gaiden both keep enemies in a single group, with one enemy breaking from the group to make a single attack. This works well with smaller groups, but for our game, we want over a dozen enemies at once, so we need to spread them out more if we're going to fit everyone.

Mark of Kri and Genji felt a bit artificial, because one enemy from the group would call the player character out for multiple attacks while the others watched. Genji could at least argue that the pattern fit the dueling style of the game. I did like the surprise attacks that Mark of Kri added from the far group-it made those distant enemies much more meaningful. The player has to keep half an eye on the outer ring at all times.

But overall, God of War and Heavenly Sword had the best feel. Having multiple enemies near you keeps things on edge and makes it harder for the player to tell what to expect next-which was reassuring, since that was the basic direction we were already considering.

Article Start Page 1 of 6 Next

Related Jobs

PuzzleNation — Norwalk, Connecticut, United States

Senior Mobile Engineer, 100% Telecommute
4A Games
4A Games — Sliema, Malta

Lead Game Systems Designer New IP (Malta)
Sucker Punch Productions
Sucker Punch Productions — Bellevue, Washington, United States

Sound Designer
Sucker Punch Productions
Sucker Punch Productions — Bellevue, Washington, United States

Technical Sound Designer

Loading Comments

loader image