What are you working on?

[qbism]

A cool AI system. A couple ideas-
Multiple prerecorded paths, bot picks best one based on conditions or when off-course. Somehow it picks... don't know how... triggers? Can aim be separated from movement?

Better nodes instead of pre-recorded paths. Migrating from AceBot to JAbot would be an improvement. JAbot has jump movements and movetypes. I don't know if strafejump is one but perhaps could be added. There are some obvious disadvantages to nodes, but one big advantage is move logic separate from aiming. For example, the 'definitive frikbot waypack' includes a rocketjump movetype. After the jump, bots can aim while in the air (and also adjust course to avoid collisions? not sure...)

[jitspoe]

My original plan was something like this:

- Constantly observe player inputs/movements and record paths. Discard small paths (ex: strafing back and forth).
- Recorded paths would have a start pos, end pos, and time so they could be chained together and plugged into a pathfinding algorithm, like A*. I'd do some cast to see if it was possible to get from the end of one path to the start of another.
- Hard limit on the number of paths (a couple thousand?) to prevent popular maps from generating huge files. New paths would attempt to replace similar paths (approximately the same start/end and time), or just randomly replace paths (ideally, weighted to replace less frequently used paths).

The problem with aiming and strafe jumping is that, in order to maintain an optimal strafe jump acceleration, you have to turn while you jump. You can, of course, maintain current velocity and rotate freely, or strafe to the side or backward to maintain acceleration, but this will cause you to deviate from the path.

I think, instead of recording raw input, I need to record positions and velocities of points along the paths, and have the bots be "smart" enough to accelerate as close as possible to these points.

[hogsy]

As usual, nothing major to show off. I've cleaned up all the directories for OpenKatana over the past few days and made some modifications to the file system so that the base path for the game data isn't hard-coded anymore and is instead the given base path is given by a script located in engine/scripts.

[jitspoe]

I don't really have high hopes for a lot of practical applications of genetic algorithms and neural nets in the realm of Quake AI, but they do look fun to play around with, so I'm going to try making a little 2D "game" for AI to wander around in and experiment.

Right now I'm just getting the data structures set up for a net:


Randomly generated neurons, and their connections (red ones are firing).

The plan is to throw a bunch of these nets at a task (after I set up the "sensory" and "motor" neurons), mix and match the best performers each generation, add some mutations (additional neurons, different connections, etc), and repeat until hopefully something cool happens... maybe...

[jim]

jitspoe: Looks like the pentagram thing from Doom3...


I textured/tweaked my player model, made some energy pack model, tried some of my earlier textures with few new ones... basically got bored of my earlier stuff during christmas. Going for some lowres style now, seems pretty fun (no idea why I ever stopped it some year/two ago).

[ceriux]

jim which map format is that?

[jim]

Q1BSP with colored lights. Made the level with JackHammer, and compiled with hmap2.exe. The green light has a lightstyle

I've also worked on a 256 color palette, but then got undecided about brown colors, how many shades for each color, should I have 16 shades or more/less? I think I won't use the fullbrights. The _glow textures feel better with any color being able to glow. And maybe I don't use the shirt/pants color thing either, and just make the team colors with different textures. I guess all the previous textures change colors a little when I get the palette done.

[ceriux]

It looks really nice for being regular q1bsp. The lighting really helps I guess.

[Seven]

Hello jim,
I love to watch your screenshots.
Your project looks really great and promising.

[frag.machine]

Time for some screenshots from my start.bsp remake... Which evolved to a start in a whole new Quake episode.
Pay no attention in the ugly skin in the fullbright bell model (DP light bug is driving me crazy):

[Spirit]

That would look so much better with a texture that did not tile as much and/or has less contrast in it!

[jim]

Think it looks like something that could be in Hexen.

[Seven]

Hello frag.machine,

Just when I almost thought you´ve forgot about the 'start' map, you post the screens
I like maps with new and interesting reliefs / portraits which are modeled into walls.
You make good use of these.
Really nice ambience !

Will you make a version as original ID1 start map replacement ?
Cause you mentioned a new episode...
It would be great if you could do a id1 replacement (like you once said).

By the way, I recently stumbled upon a worthy oponent for you frag.machine:
vs

[drm_wayne]

Finished my engine based class / teamselector (for deathmatch only) using some new custom cvars :

[frag.machine]

Will you make a version as original ID1 start map replacement ?
Cause you mentioned a new episode...
It would be great if you could do a id1 replacement (like you once said).

Well as I said this map started as a remake of the original start, but at some point I began working in another maps and added some custom QC (I do this a lot when I map, I rarely map to just vanilla Quake, I need to add custom entities and effects) and then an idea for a mod emerged and suddenly that unfinished start map looked the exact entrance door to my mod/episode thing, so it's very unlikely it will turn to be just a replacement to the regular start.bsp. Sorry if I disappointed you.