Chade on 29/7/2009 at 04:38

---

That's not lazy, that's perfectly normal!

I'm not expecting you to post detailed blueprints of all your ideas. Unless of course you want to, or that's the way the conversation flows. But don't feel like you have to go to great lengths to "convince" me of anything. It's not really possible: the only way to get serious about this stuff is to actually prototype and test ideas. Anything else is just armchair logic.

Chade on 31/7/2009 at 02:22

---

I have been thinking a little bit about procedural guard patrols lately. Not much, but enough to wonder if it might actually be possible to generate reasonable procedural patrol routes.

A few issues spring to mind:

How do you model the map?
How do you generate a set of possible patrol routes around that map?
What metrics do you use to score those patrol routes?

(My proposals here certainly aren't good enough for a real thief game, but they might be a decent starting point. I don't think about sound, or the players perception, or how the level works if a guard is alerted. All my patrol routes are completely deterministic.)

~~ Modelling the map ~~

My first approach would model the level as a discrete set of nodes, and paths between those nodes. I assume a node in space is either "safe" or "not safe", as a function of time. Nodes in shadow are always safe. Nodes in light are safe if no guard can see them. Each pathway between nodes tells you how long it takes a guard to travel that path, and what nodes are visible to the guard as he travels that path. Given a set of patrol routes, you can then calculate whether any node is safe or not at any point in time. because we are limiting ourselves to deterministic patrol routes, this function repeats itself, so if we restrict ourselves to medium/small patrol routes, then the overall behaviour of the mission stays constant over time.

~~ What possible patrol route sets do you consider? ~~

No idea. I haven't thought about this yet. Presumably it has to look like the guard is "going somewhere", or has "some purpose". You want guards standing still at the front door. You don't want guards standing still at the attic window where the player will make his "stealthy" entrance. This sort of thing can't be captured by the model I outlined above. Something else has to provide the architectural meaning that constrains the possible patrol routes.

~~ What metrics do you use to score patrol routes? ~~

I have no definitive thoughts on this yet. But here are some metrics which might be usefull ...

- You can calculate how long the safe periods for a given node are. This is a pretty obvious metric, but also probably one of the most usefull ones. The more nodes with short safe periods, the more tense the level.

- Given a pathway between nodes, you can calculate whether the nodes safe periods overlap. This gives the player the oppurtunity to safely move along that pathway. The amount of overlap approximates how dangerous the transition is.

- Some nodes can have "trap" periods ... these are periods which are safe, but will turn unsafe in the future without any of the neighbouring nodes turning safe. A player in a trap period will definately be spotted in the future, but gets a bit of time to plan what to do about it.

- You can calculate the shorted path between any two nodes. Then you can compare it to the shorted *safe* path between those nodes (if desired, you can do this multiple times with varying danger thresholds, which the player will not exceed). The ratio of the two tells you how indirectly the player must move to safely travel between those nodes. If the player can just get directly to where he wants to go, he won't feel like he is sneaking. But you probably don't want to make the route too indirect either.

- I'm sure you can think of many more metrics!

(As an aside, these metrics would not just help score patrol routes. You can also use it to procedurally choose where to place loot, for instance. Or choose whether to add shadow to some nodes.)

Albert on 3/8/2009 at 05:14

---

What Garrett needs is a Bush disguise... And I don't mean Bush, That would get him killed quicker. ;) No, I literally mean just any disguise so he can blend into his environment. I mean, some games have done it already, but with Thi4f (Shudder, bad l33t sp3ak), that would be an unnecessary but experience-keeping addition.

Beleg Cúthalion on 3/8/2009 at 06:17

---

The patrol system is quite complex already in TDS/T3Ed. You can have branching routes with different chances for each and with a little scripting you can move any AI to a completely new patrol once alerted etc.. TDS has a lot of paranoia scripting as well but I fear it's again too subtle to notice. Another chance for FMs to make something of it.

Beleg Cúthalion on 3/8/2009 at 20:06

---

There are quite some of them still in the missions although they're hard to notice. It took me some time to re-connect this patrol when I combined the two PaganIntro maps for the T3 Gold project:
(http://s12.directupload.net/file/d/1875/vln8pf33_jpg.htm) http://s12.directupload.net/images/090803/temp/vln8pf33.jpg
(right in the lower corner, the numbers indicate chances, arrows link patrol points, pink lines point to look and animation points)

All the time this is just a Pagan walking between old cupboards.

I cannot say anything useful about the paranoia things yet because they appear to be single properties/things/whatever which don't tell their secrets easily. I'm thinking about doing some experiments with stuff like that anyway so maybe after a long long time you can see some of it. :p