Gijs-Jan

Which specific parts?

Well, I guess I have to dumb it down a bit then . The point is that the AI also uses quite some statistics to determine where you *might* be. Seeing as how an AI is sometimes spread across the entire map (especially with alien base, terror missions and the like), it has quite some information on where your squad *isn't*; and thus, also *is*. Yeah.. the AI doesn't really care that much if it kills itself, if the benefits are high enough. (Roughly 2~3 to 1 ratios; depending also on how more valued your unit is (9 rookies ~= 1 Colonel) This should be fixed soon; TUs should be spent when teleporting, and the AI shouldn't do the Veni-Vidi-Vici tactic with teleports anymore.

Yeah, it was a decision that Defensive aliens stay inside the ship at all times. Which we'll probably revert now then

I'll have to fix that then

The Reapers AI is disabled in this build because of a bug in the zombification process whenever more than 2 victims were hit in a single turn. We needed some time to work it out as fixing it wasn't trivial. They should be in the next build. As for grenades; it's strange that the grenade got thrown a distance of 15 tiles. Do you maybe have a save of that game? Aliens running out of ammo should also be fixed in the next release.

It's already fixed (obviously) Chris was paying attention. You can probably blame it on the bad weather that's been terrorizing western europe.

No problem, I thought I fixed the behavior some time ago; so no worries if he's working with an earlier version :-)

*Proud moment for me*

Do you have a save of that? Quickly, before Chris notices!

Well.. At the very least it shouldn't be running towards the aliens! Do you maybe have a save before?

First off, thanks for the discussion. You guys already found quite some obscure bugs for me to squash. The 180 degree reaction fire is a confirmed bug now; and it'll probably be gone soon. The 360 degree turn on the spot is legacy behaviour; it's purely cosmetic/burning APs. I thought that you guys rather see more complex maneuvers than realistic pirouetting behaviour; so aliens that move / are in their turn observe in a 360 degree circle. If you consider this to be cheating; I can obviously just manually make them pirouette, but in my opinion it just looked silly to see each AI agent turn on the spot before starting to walk I'll also take a look at the ignoring of floors, which might be a bit harder to solve; because there are no real constraints from the AI perspective; so I suspect its something in the groundcombat code that's holding them back. Goals this summer: - Far more realistic search behaviour. - More diverse behaviour per race (and more correct usage of abilities) - Better defensive behaviour around UFO's. - Better coordination, in team and overall. (expect backup etc) - Overall better tactical engagement.

Keep the theories coming! At the very least it gives me a general direction to aim at when going bug hunting ;-)

So.. I think I have to do a mea culpa for both the sight bug and the overload on TUs! The sight bug was a combination of two things: - A small bug in the AI in which the engagement distance was used for detection instead of the actual maximum sight. - The introduction of a probabilistic system which tries to estimate where your squad members are; which was set far too aggressive. The overload on TU's was due to me forgetting to remove a debug line which gave units on the map some TU's for a movement scenario I was testing. The scenario was removed; but the circumstances under which the units gained TU's wasn't. All should be well again in the upcoming patch. :-)

Seems I forgot to post my answer to Hawk66's question, my apologies! <Warning, bit more technical than normal> The units will be able to act as a squad (If all goes well, in the very-very near future). And to be clear: the planning logic will use a custom HTN (Hierarchical Task Network) logic planner. Because Xenonauts is written in C++ and the planner needs to be efficient in the extreme, I can't use any quick prototyping languages. This is the main reason why it's taken this long. The HFSM is only used to alter the motivation (with fuzzy logic) for the units. So, each units describes certain characteristics (P.e. aggressiveness, etc) which are used to influence the score certain actions give. However, the planning of the actions is done by exploring a state space using HTN. The plans formed by the HTN are then evaluated using these characteristics. This, coupled with some randomization, will ensure that logic is both rational, but unpredictable. (to some extent)

Hey guys, Sorry for the lack of blog updates on my part. This is in part because I went through a multitude of techniques before settling down, part in because of the bumps on the road while implementing, and in part due to that I'm still deciding on how to present it to you guys. I don't want to get to technical, as it won't be approachable, but there are some technical parts which could be interesting for other gamedevs. For those who want to get a small insight on the technique I settled on, can read this: http://www.cgf-ai.com/docs/multi_unit_htn_with_astar.pdf or take a look at plannedassualt.com :-) As soon as I'm done with the blog update, you'll know!

I'll try to get a new blog post up soon which isn't so vague ;-) The eta on the rest of the AI I'll have up by the end of the week.

I need to correct on what I said earlier: there is a simple decision layer for some of the knowledge. But it is just that, simple. There is target prioritization and things like avoidance; the only thing is that the A.I. is not yet able to formulate plans, or foresee the consequences of actions it chooses.

I don't want to spoil too much for you guys; because I know how peeking behind the curtain can spoil the experience. While there should be some new behaviour for the new A.I. in V16 (shooting from beyond unit LOS, but in squad LOS; better running away; cover usage; psionics; etc) the "real" A.I. hasn't been added yet. To explain, I've been primarily working on adding knowledge to the A.I.: things like recognizing dangerous situations, good targets, etc. The A.I. now "knows" about these things, but has no real way yet to think on actions or formulate a plan. This planning part will be the final real system to finish. To overview: the knowledge comes first, followed by the ability to plan actions given a situation, and finishing finally with the ability to simulate the consequences of actions. That being said; I still like to hear everything you find so far and like (or not), so I can steer the design! :-)

Ah too bad... I was already looking for ways to hardcode preset difficulty levels for a certain someone..

Heightened? So there is already a baseline of aggression?

This; shots fired outside of LoS still follow all the rules with regards to normal shots. The A.I. also won't shoot if the accuracy % drops below a certain percentage.

Giovanni; As for the missing sprite, I would suggest a fallback sprite and a text overlay over the unit which lists the missing sprite. (We could make it a hotkey enabled debug option)

Best you can do is provide us with screenshots so we can look into it, and/or mention which tileset was used :-)

Could you provide us with some more information on when the crash occurs, like: - Where was the UFO crashed? - When did the crash occur: -- During start of combat / before any movement. -- During the player's turn or during the AI turn. -- During moving of units, maybe reaction fire from the AI (or player character?) - What were you doing just before the crash?

Very true! And don't forget the optimizations! If you think you're ready to curse the loading screen / hidden turn screen for displaying too long; try to step in our shoes where you need to look at the damn thing all day long!