We know how to chase the player, evade him, and keep an eye on his position. We have even predicted his future actions. So, now it's time to work on the mechanics of shooting. We need to learn when is it a good idea to shoot at the player in different contexts. We also need to know whether we are handling a machine gun, a sniper rifle, or a catapult. As you will soon see, each one requires slightly different approaches to the problem of targeting.
Before starting, I need to give some personal words of warning. The same way I enjoyed movies like Star Wars, Alien, and Saving Private Ryan, I think games with a fighting/shooting element should be recognized as enjoyable experiences. I don't have any moral problem with that, and I think any healthy person can differentiate between the fictitious violence shown by games/movies/books and real violence. On the other hand, I'd recommend that people play many different games, not all of them with a combat/violence component, just as I'd go to see different types of movies. That said, the following sections deal with the specifics of shooting, so they necessarily focus on the algorithms required to target and shoot down enemies.
The first approach we will explore is shooting with a weapon that has infinite speed, or in practical terms, very high speed compared to the speed of the target. This can be the case of a laser gun, which would advance at light speed, for example. Then, we can assume the time it takes for the projectile to reach the target is virtually zero. Thus, the selection of the shooting moment is really easy. All you have to do is make sure you are well aligned with the target at the moment of shooting. As the velocity is very high, we will have a sure hit because the target will have very little time to move and avoid the collision with the bullet. Clearly, it is not a good idea to abuse infinite-speed weapons because they can unbalance your game. If you build these weapons into the game, make sure you balance them well in terms of gameplay. For example, the firing rate can be very low, the ammunition limited, or the weapon might be really hard to get.
What happens with a real-world firing device? Even a real gun shoots projectiles at a limited speed (approximately 300-600 meters per second). This means shooting a fast moving target is harder than shooting one that stands still. Thus, most weapons must be modeled as finite-speed devices, where some careful planning is used. I will explain two popular approaches.
Version A: The Still Shooter
The still shooter targets the enemy and only shoots whenever the enemy is standing still for a certain period of time. The reason is simple. If the bullet takes one second to hit the target, and the target has been standing still for a certain period of time, it is a good hypothesis to assume the target will stand still for another second, thus making it a good moment to attempt shooting.
An enhancement to this algorithm is to watch the target for specific actions that indicate restrictions in his ability to move. For example, if the target is standing still, he might begin walking in any given moment, thus making it an unsafe target. But what happens if he sits down or if he is tying one of his shoes? Clearly, we have a better aim here because we know for sure he won't be going anywhere in the next few seconds. This would be the kind of reasoning that would drive a sniper-style AI. He looks for very safe shoots that hit the target most of the time. By shooting only when a safe hit is granted, the shooter ensures one kill while not giving away his position easily. The disadvantage is that maybe the shooter will have very few opportunities to actually shoot, so it is a good idea to make him less restrictive. The way to do this is to introduce errors in his processing. He might sense time incorrectly, confuse animations, and so on. So sometimes he will shoot when he's not supposed to. When done carefully, this can accurately model fatigue and morale, affecting the ability of the sniper to stay focused.
As a summary, here is the algorithm in detail:
Global variables: Timestill time since the enemy began standing still StandingStill 1 if standing still, 0 otherwise When it begins standing still StandingStill=1 Timestill=now If StandingStill and more than X seconds have elapsed since Timestill Shoot
Version B: The Tracker
The tracker AI also tries to model the behavior of a sniper. In this case, he will shoot moving targets, not just those who are standing still. Shooting a moving target is really hard. We need to combine the shooting behavior with a target tracking routine, and there is a predictive component going on as well. If the gun has a finite speed, we need to target not the current position, but the position where the target will be when the bullet hits him.
The idea is simple: Compute the distance from the sniper to the target, use the projectile velocity to compute how long it will take for the projectile to reach the target, and predict where the target will be in the future, exactly when the projectile arrives. This way you can aim at that spot and get a safer shoot, especially in distant or fast-moving targets. The algorithm in full is depicted in Figure 7.7.
float d=distance (sniper, target) float time=d/bulletspeed point pos=predictposition(target,time) if aiming at pos shoot() else target at pos;
Whether predictive or still shooters, we have focused so far on single-shot firing devices, where each shot is considered an individual AI decision. But other weapons, such as machine guns, offer the possibility of shooting bursts of bullets at high frequency but with reduced precision. The AI logic for such weapons is a completely different subject, and thus deserves its own separate discussion.
Machine guns offer fast firing rates at the cost of inferior precision. Shots cause the cannon to shake due to recoil, making it hard to aim accurately. Thus, their main use is found not in targeting people, but areas. The machine gun is aimed in the right direction, and short bursts are fired to hit anyone in the area.
The first type of gun we will examine is the fixed machine gun. This kind of behavior is exhibited by gunners in bunkers, trenches, and so on. Some classic guns would be the MG-42 used by the German army in World War II, the M60 used in Vietnam, and so on. Here are some stats from the former:
From these statistics, several lessons can be extracted. First, these guns hardly ever moved, but instead kept on targeting and shooting down enemies from a fixed position. Second, these guns did not have a lot of autonomy, the standard feed type for the MG-42 was a 50/250 metal belt. Thus, a burst could not last longer than 10 seconds, followed by a pause to change the metal belt. These guns were thus used for performing short firing bursts. Their algorithm is relatively straightforward. By default, the soldier stands still, waiting for new enemies to arrive. Then, as they begin to get closer, the gunner must rotate the gun to face the enemy. Rotation must somehow be penalized for slower models. When the angular difference between the gunner and the enemy is smaller than a certain threshold, the gunner will hold down the trigger while trying to refine his aiming. Keep in mind each shot introduces some distortion to the aiming due to recoil, so the gunner must re-aim every time. As a result, fixed gunners do not usually aim carefully; they aim at an area. Thus, these gunners are especially useful when we need to stop a wave composed of many soldiers. By pure chance, some bullets shot by the gunner will reach their target.
A common mistake is to forget about feed sizes. Many World War II games display machine guns that seem to have infinite ammunition.
Let's now examine the problem of a moving character carrying a light machine gun, such as an AK-47 or an M-16. As a rule of thumb, only movie characters use moving machine guns to shoot long bursts. Recoil makes it impossible to aim, especially if standing up. So, ammunition is wasted because most projectiles will be lost. Besides, these guns do not have long cartridges, so ammunition must be used with care. Here are some stats from the World War II Thompson submachine gun, aka the "Tommy gun":
The gun came with 30 bullet cartridges, and a soldier in World War II usually carried three such magazines. As you can see, ammunition was still more of an issue than with heavy, fixed machine guns. Thus, the most common practice is to treat these assault guns as rifles with very high firing rates. Bullets are shot one by one or in very short bursts. The only situation where a moving gunner can effectively waste ammo is in a fantasy setting, such as space ship games. Here we can forget about realism and make the tie fighter or other ship of your choice shoot long firing bursts.