If the bad guy is in room 3, and you are in room 1, then you can’t see the bad guy, and he can’t see you. However, since it’s all modeled in the computer, it knows where you are, and thus can tell the bad guy “He’s in room 1, go get him”. This is cheating at the worst level. The computer has more information than you do, so it becomes unbeatable. The challenge is to limit what the bad guy can see and hear, so he can only respond to that stimulus.

So given these conditions, it isn’t that hard to write code to deal with the decisions a bad guy can do. If he sees you, he shoots at you, if you shoot back, he decides to either run away, duke it out, or find some cover to shoot at you from. What happens if you run away? Well, lets make him follow you. Now we run into the next big problem. Path following. If, as in slide 3, I meet up with this guy in room 1, and I run from him into room 3, how does he track me? Doing this in a convincing manner is the hardest thing currently in games to get right. There are a number of approaches to doing this, some more expensive in processor time than others, and some more accurate than others. Here a couple of them.

Map markers. This is the system we used in Heretic II. Markers were placed in the map all within line of site of each other. Bad guys followed a direct line from one marker to the next, thereby navigating the world. This also has the advantage of allowing information regarding the world, such as doors, holes and so on to be embedded within the markers, allowing the bad guys to leap chasms, and open doors at will. The downside is that these markers have to be placed in each map that bad guys inhabit.

Way point. Similar to the marker system, the bad guys themselves drop markers in the world every 10 seconds, to create a kind of ‘bread crumbs’ path. The down side here is that initial forays into unseen area’s mean some wall bumping before there are enough way points to successfully navigate the world. Plus, with each bad guy dropping these markers, you seen end up with a ton of them.

The last method is the most successful, which involves real world cognizance. IE doing traces in the world to find out where walls are, where the floor is, what obstacles are in the way and so on, and making real choices about what to do next. This is EXTREMELY expensive on processor time, but is the only ‘real’ solution to the problems of path finding. We do have a new system we are working on, but I can’t go into details at this time, mainly because no one has explained it to me yet.