The Ur-Quan Masters Discussion Forum

Gemini tested the hypermelee in SCII to see if the points values are correct. Assuming that's close enough to UQM to post here.

https://www.youtube.com/watch?v=NO0Ja1iZms0

i thought shiver already figured this out?

Well actually, not all of what Gemini does is the first time it's been done: I think he does it for the sport. If there's a thread where shiver posted something similar, I couldn't find it just now. If you know where it is, I'm sure it'd be useful to compare.

https://wiki.uqm.stack.nl/User:Shiver/Balance_Mod/List_of_Changes

Right, well these works, believe it or not, can actually completent each other. I don't see anything about AI vs AI in what shiver did, and also he's only indirectly answered the question "how much should each ship be worth" since he's implemented tons of changes as well. If there's anything like the table gemini compiled, I don't see it in what you linked.

Shh! This video is super interesting. I also don't recall ever trying to map out the odds of every match-up in PvP, which would really be more of a "Shiver vs Gekko" map if such a thing were ever attempted.

uh oh...i meant no disrespect...the more the merrier i say.

He says the AI's defects won't have a big impact… but it definitely flies some ships well and some very, very poorly (Kohr-Ah, Earthling). I haven't watched the whole video to see if he notes this. I'll get to it…

Not to mention that the AI's defects are exactly what makes some ships (like Spathi) so crazily effective against the AI.

Yeah. But my real complaint isn't that the AI is bad, it's that he took his results as solid anyway. For isntance, he called out the Supox as 'actually' being weak right after showing how the AI was flying it in a completely ludicrously stupid way.

I wonder if we could do a PVP version of this with enough dedicated netplay…

Has there ever been any attempt(s) to write more sophisticated AI player code?


I think it might take a more multiplayer-active community.  It is not just the total hours needed to play every possible match up so many times for a decent sample size.  You would need to only consider the results of battles between top players. Further, you would need many dozens of consistent online players to know that your top players are experienced and well rounded in their play styles.

We had a playerbase for net melee for a few years. It was less than dozens playing at all hours, and it was over than a decade ago, but we had one.

I would worry that statistics based off games between net melee experts still wouldn't be that helpful. The top players all use Utwig as a Kohr-Ah counter, but that match-up doesn't go anywhere unless the Utwig player's blocking game is on point. It's a lot harder to pull off against a Kohr-Ah player, even a shoddy one, than the cyborg. We could list "Utwig hard-counters Kohr-Ah" using empirical evidence and the lower 50% of players might never be able to replicate that.

Yeah, it's complicated.

So then what about using an improved AI for mass testing?

Are there any community-developed improved "cyborg" players for UQM?

Writing AI is said to be one of the most difficult parts of programming. There hasn't been much work done in this area by the fan community. I did a little for Project 6014 to get the Lurg Prawn to fight and made minor fixes to some of the core ships; e.g. stop Ur-Quan from launching fighters at cloaked Ilwrath. Anyone feeling motivated to write some AI is welcome to jump in and give it a try. A good place to start: Turn VUX into an actual threat for the campaign.

Glancing at the code... it may be that some of the difficulty is actually coming from the software architecture and documentation rather than the inherent complexity of what a UQM AI must handle.  I mean for one thing, you cannot just throw all these bitwise operators at us simple, gentle millennials! :D

IMHO, as much as possible, you want to treat ship AI scripts as content; as separate from the engine code. Ideally, it should be easy for your AI scripter to know what ship data exists that he can get access to (and easily access it) plus what "commands" the AI can give to its ship directly and to widely-useful subroutines like "turn the ship towards the direction where a shot from my own weapon #1 would intercept the enemy ship, given the former's speed and the latter's velocity".

I see there is a Lua branch of UQM on sourceforge, which presumably is intended for making the game more modder friendly.  Unfortunately it has no activity since 2013.



Is that possible when the player has cheap, perennial Cruiser production?  Against the Cruiser, is the Intruder good for more than just nuclear target practice in network play?

Er no, not really, come to think of it. There's a way to play that will let VUX win once in a while, with no help from the VUX warp-in ability. The laser can shoot down nukes with some skill, and you can get the Intruder drifting around looking for the planet while trying to defend yourself. After a lot of stalling, slingshot off the planet to chase down Earthling. Which they can still avoid, sometimes entirely, and sometimes they'll merely be grazed by the laser before slipping or bouncing away.

It would take ship buffs or some AI tuning with ship buffs to make VUX fit in nicely. The goal wouldn't be to make VUX deadly, just make it a fight. Even Umgah is a fight.

Yeah, that's a rough one. Maybe let them escape and re-intrude? But the cruiser captain would still be able to just keep moving.

Make the laser quickly raster back and forth in a narrow cone? That'd help with shooting down nukes.

And that still assumes that the AI is smart enough to pull off the planet search strategy… but it doesn't need to be as good at it.

How many ships in total would need these AI and/or physical behavior changes? How many ships are already well balanced and played competently by the AI?
I do not want to discourage piecemeal hacking, but the broader and deeper the desired changes to ship combat, the more efficient it becomes to first modernize the software architecture of the ship combat code so that these modifications become quite a lot easier to make.

Are there already plans for refactoring ship physics and AI into an Entity-Component-System model and/or moving ship data and AI behavior scripts out of the C files?



Unless the nukes are slipping in "between" the ship facing directions or avoiding the hit-scan detection by crossing the laser beam between frames, the AI might better benefit from a simple improvement of the intruder's turning speed, combined with the AI making nukes a higher priority target and/or better anticipating their movement.  The nukes do accelerate and turn a bit, but they are predictable about where they want to go.


I do not know how the "wrapping" mechanic affects the planet position, but otherwise it should be fairly simple vector math to engage in a random gravity whip maneuver.  Where it gets complicated is trying to aim for the enemy ship using the gravity whip.  There's a ton of variables to getting that remotely right.

If we're just talking AI tweaks for the campaign, not too many ships...

Umgah could stand to use its cone for protection against long range missiles a bit better. They let some easily-stopped attacks slip through. It's otherwise a great opponent.

Ur-Quan fighters should be less of a liability. Give them collision proofing against asteroids and enemy ships, and longer fuel reserves to return to the Dreadnought with. They don't need to be immune to attrition tactics, but they shouldn't be going kersplat left and right so much. No AI changes needed.

Kohr-Ah doesn't need to change, yet the amount of wasted potential with this ship is staggering. Why, why does it run straight after everything it fights? It should behave more dynamically. I say that, yet they could very easily become way too difficult for most players with the right (i.e. wrong) playstyle. This would a be a low-priority ship to work on.

Orz AI is terrible, but fixing it isn't important since fights against them are optional and comparatively uncommon.

Thraddash could use just a bit more rate-of-fire and crew or something like that, no AI changes needed for them.

And VUX (already discussed) is the one real offender in the midst of this.



I wouldn't know what that means. It's doubtful anyone has a project like that in the works, core team or otherwise.