[ Closed ] [SUGGESTION] How to make the bots more intelligent ? (Page 2 of 5)

I think there should be a target practice mode where the bots move either left and right or up and down.

I've always been a fan of finite state machines. Of course, it would be possible to tweak it into something more complex -- or to simply stuff them all together into one loop. I think a finite state machine would work better in this instance because of how it blocks everything out nicely. There are always hooks back to a "core" behavior.

As for pathfinding, I'm only familiar with A*, but I know there are numerous other ways of doing it. I know for my autonomous critters I've created, I generally make them stupid: they move until they run into something, at which point they try to go around it. You can see a slightly more intelligent version of this here: http://www.red3d.com/cwr/steer/Obstacle.html and http://www.red3d.com/cwr/steer/Unaligned.html

I think a hook calculation would be relatively easy to factor in. We'd need formulas for how the hook extends based on distance and velocity, but from that it's a pretty trivial calculation to figure out where we're going to end up (retraction of it is based on velocity, so at a high enough speed the retraction factor is small enough to be negligible, and at other speeds it could be factored in, but it might not have to be). A possible way of programming this behavior is that if a bot chooses a target above itself, it looks to see if there's an edge nearby. If there is, plot a path at puts the bot within hook range of the edge, hook it, and spin around, then continue path. Ideally, this sort of behavior would be exhibited naturally by describing ranges of motion and allowing the bot to string them together.

One also has to factor in which weapon to use. Generally speaking, the hammer should be used only as a last resort (if all ammo is gone), with the one exception of all ammo gone except for the rocket. This way we wouldn't have to worry about the bot blowing itself up. Of course, if the bot is given a short term memory of sorts, it could take a gamble -- if it thinks the target is weak enough and the bot is healthy enough, shoot a rocket at the target in close range. If the target is within a certain range, prefer the shotgun to the pistol, especially if the target and/or the bot is moving. If one or both are stationary, the rocket may be a better choice.

@nikoPSK, clasic_traveller_diehard
If I am not mistaken, you want (even dumb) bots in order to train at aiming.
I think that this kind of bot won't permit a good training (or maybe for very beginners) if you can anticipate their moves and if you don't have to avoid projectiles.
Besides, I am much more interested in the creation of "clever" bots than brainless moving targets. Of course the first versions could be used as you describe, but it should not be the final objective.
What do you think ?

@GSD
We definitely have different approaches, but that's interesting ! I like not to tell the computer exactly what to do, but just give a hint and let it find out. For example, considering the weapon choice, once the bot knows how to use each one, I was thinking of doing the following (roughly):
- Choose some relevant (or not) informations (target's proximity, target's speed, target's direction, target's health estimation, edges between me and the target, the bot's munitions and skill for each weapon, the bot's health, the bot's speed and direction)
- Create processes taking this information in parameter like Neural Networks or with a fuzzy logic system that the bot could tune by changing some coefficients.
- Let the bots compete with reinforcement learning and natural selection, first with each other, and then online

The main advantage of this approach is that you could, with no efforts, end with a lot of several bots that players could not get used to (or bored of).

The major problem I see with a neural net is that it takes time to train it. Players don't want to wait for their bots to "learn" how to play -- they just want them to get up and _go._

I suppose it would be possible to ship a certain number of pre-defined bots, and let the players grow their own, as it were. I'm also wary of how a neural net might get confused, or latch on to a behavior which is a poor idea. Still, some form of individuality in the bots, whether through a genetic algorithm, neural nets, fuzzy logic, or any other piece of AI jargon you want to throw up, is always a good thing.

@CLASIC_TRAVELLER_DIEHARD
Either I did not get it or it is off-topic in my opinion.

@GSD
I was not thinking of releasing blank bots

@raph
What do you mean by "bot coder" ?
Anyway you're right about the fact that the first thing is to have a bot, even dumb, than can interact with environment.

Does anyone had time to study the source and found something relevant ? I did not have time to try anything yet...

let me explain
a "landmine" is a bot programed to sit around and swing a hammer and is then spawned in inconvenient locations (such as the bottom of a pit) with the hope of a tee landing on it it also has the advantage of being move able(via hook) and destructable

a "building block" is is just the dummies matricks described spawned in stacked formations (example: pyramid) and through out the battle the structure would be destroyed or the block harvested by players for their own fortifications

my name is NOT clasic_traveller_die its CLASIC_TRAVELLER_DIEHARD classic traveller is a si-fi rpg from the same era as dungens and dragons lookit up on wikipedia

This is indeed a challenging aspect. And my main concept would involve a Ray-Tracer in mutiple directions. Tracing not only from within the current position, but rather from several units above us ,too. Looping through all current enemys on the screen, evaluateing possibilities of pursuit or retreat..yeah that could be fun. I [i]could[i] contribute parts of my client-side aimbot, rocket-bot and autohook, though i have written this in Delphi. But then again i see how this can be dangerously abused, so i'll better pass on this one.

I saw plannings of neural networks and implementation of very complex AIs here. Is it that necessary? Well, someone could write a bachelor thesis about this doing this properly.
I guess it is a bit overkill for this game and would be only good for a proof of concept, but hey, you never know what comes around in these projects.

Anyone seriously on to this?

GSD: I would recommend you get crackin'. We don't exactly have a closet full of spare C++ coders around, so the more, the merrier.

I'm in the US -- CST (GMT - 6). Alas, my days are only going to get busier for a while, but I'll still do what I can. I'll start working on some C++ stuff this weekend.

From a perspective of hacking things, there's a _potential_ it could be abused, but I think if it's coded properly we wouldn't have to worry about it. Simply sending commands to a server saying "shoot with rocket" or "aim hook here" won't be enormously useful to people.

It might be better to set up a Google Code project, because of all the tools it gives us.

I meant we should set up a Google Code project instead of creating an entirely new forum. If he meant on a forum on the Teewars site, that might work, but Google Code has the central repository, forums/news feeds, that sort of thing. The hosted projects are designed to house these sorts of collaborative projects.

Why do u need bots?
Play against real players!

So what's our next move ?
Should we plug into the teewars trac ? how ?