Incline Remnants of the Precursors - Merry Christmas, bitches

[RayF]

It's a good exercise in game development I'd say as you already know how the end product needs to behave & feel, which prevents your scope from going way overboard and getting lost somewhere on the way because shit got out of hands.

Yeah, scope creep really seems to kill a lot of these projects. As much as I'd like to see it complete successfully, M.O.R.E. is a textbook example of scope creep.

Another real problem is that a lot of indie developers (i.e. most of them), including me, are NOT game designers. As a result, they think of things from a player's perspective and often throw in a lot of features that look good on paper but end up making the game a huge mess.

I don't claim to be a game designer so I am keeping it simple and sticking to what may have been the best 4x formula made -- MOO1. Once that is done, I'll then experiment with things I think are improvements but always keep the original ruleset intact.

 

[Luka-boy]

But from a "gameplay" standpoint, it should play much better -- the AI should be smarter, exploits will be closed, the extremely rare micro issues are squashed, and the UI should be much easier to use.

Oh god.

Will we believe a golden Bulrathi can fly?

 

[RayF]

word. "Conquer the stars" is a perversion.

I'll answer the AI question when I have more time to compose a thoughtful reply, but I would like to point out that MOO4 has never been a reboot of MOO1. The game mechanics are very clearly based on MOO2; they are just missing a few features: Antarans, heroes and the 3 add'l races. Those will arrive with DLCs most likely.

Also, the MOO4 tech tree is a Civ-style tree. It's nothing like MOO1 or MOO2.

And starlanes.

MOO4, just like the 3 that game before it, is not an iteration of a previous game but a new game completely.

 

[RayF]

I'd like to hear more about what is being done here.

ok, the two most important things to know about the AI are:
1) the game and interaction between empires is constructed so that the player is on equal footing with the AI
2) the code will be open-sourced, allowing for continued refinements to the AI by the player community after release

There are two goals in the development of a game AI (imo):
1) Don't do stupid things
2) Do smart things

The most important thing, by far, is #1. No matter how smart you make the AI (#2), any stupid things it does because the developer failed at #1 will break immersion.

I've said this before, and it's worth saying again: the vast majority of players don't want play against a good AI, they want the illusion of playing against a good AI.

Because, to be very clear, a good AI will thrash players and players don't want to get thrashed. They want to win, but they don't want to win against a stupid AI. But as soon as the AI does something clearly dumb, any satisfaction the player gets from winning will be lost.

So the hope is to meet goal #1 (don't do stupid things) while making the AI interesting to play against (personality, strategic differences, etc). This alone will make the AI better than in MOO1. And don't forget that an AI that doesn't do stupid things will generally beat a player who does do stupid things, so it won't be a pushover. However, a good player should be able to beat it on a strategic and tactical level.

If someone wants to make a hard-ass AI that beats all comers, they are more than welcome to buff the various AI classes. Perhaps I can turn the AI into a separate component so that players can swap out the AI logic more easily.

 

[kyrub]

I've said this before, and it's worth saying again: the vast majority of players don't want play against a good AI, they want the illusion of playing against a good AI.

I always wonder if people mean this opinion seriously. The other possibility is more probable: I am an obsolete gamer. Everybody else just wants to be tricked and "entertained".

There's an important difference between giving the AI bonuses (no problem with this) and teaching the AI to do smart moves (important as hell). If the AI cannot use a clever ship design, how is the game meant to be entertaining? There is a vast part of the player community that appreciates surprising AI moves, that's what games like GalCiv have taught us. To "not do the stupid things" is not gonna break it.

Perhaps I can turn the AI into a separate component so that players can swap out the AI logic more easily.

Some hope at least? You should seriously consider doing it. Bad AI can break the games popularity completely.

 

[RayF]

I always wonder if people mean this opinion seriously. The other possibility is more probable: I am an obsolete gamer. Everybody else just wants to be tricked and "entertained".

You are not obsolete, you are just part of a small minority. Didn't Brad Wardell say once that over 90% of GalCiv players never leave the easiest level? They're not looking for a strategic challenge; they just want to be entertained for a few hours. Not everyone plays games for the same reason.

The code in ROTP is going to be open-sourced specifically to allow people to mod it to their personal tastes.

 

[Destroid]

You are not obsolete, you are just part of a small minority. Didn't Brad Wardell say once that over 90% of GalCiv players never leave the easiest level? They're not looking for a strategic challenge; they just want to be entertained for a few hours. Not everyone plays games for the same reason.

Presumably that figure is provided by legitimate customers, fools easily parted with their coin.

 

[Infinitum]

I always wonder if people mean this opinion seriously. The other possibility is more probable: I am an obsolete gamer. Everybody else just wants to be tricked and "entertained".

Taking this to the current extreme; would you be willing to play chess exclusively against the top difficulty of one of the modern engines? A game on equal terms relies on the opposing number actually doing mistakes for you to exploit.

 

[RayF]

I think most people here are frustrated by the strategy genre's crutch. AI players are perfect at micromanaging to optimize for simple objectives but not so good at complex strategies like deception etc that a human player would do. Since they can't design AI players intelligent enough to navigate complex mechanics and systems, they give AI players bonuses on higher difficulty settings to compensate. Or worse, they let the AI cheat e.g. knowing more of the game state than they should, seeing in the fog of war etc. This feels very artificial and incents the player to cheese the AI instead of playing normally. From point 1 it seems that you're not on board with this approach. Good. But this doesn't say much about how you're going to improve on the AI itself!

Fair point! My current goal for the AI during the alphas is for it to effectively use the various features that are also available to the player: managing colonies, exploring the galaxy, colonizing, shuttling transports, designing ships, deciding where to send ships, deciding what to research, handling diplomacy (peace/war/trade, etc), managing spies, etc etc

If I can do all of that without making obvious mistakes (this most often shows up in diplomacy), then that's pretty good.

Let's try to unpack statement 1. AI shouldn't do "stupid things" (e.g. leaving colony ships undefended) but is there room for error or randomness? If AI always plays optimally, is it too predictable?

There is room for personality. "Not doing stupid things" simply means acting realistically, not necessarily optimally. This means that different leader types may research differently, design ships differently, etc etc. Maybe the Mrrshan will build weapon-heavy ships even if defensive components might make the most sense at times. Maybe the Psilons will hunker down behind their planetary shields and focus on research when perhaps the best move would be to expand.

"Doing smart things" means (to me) finding the optimal path. I am not as interested in this because it leads to the AI trying to optimize things based on the game's ruleset, which creates conformity as all of the AIs look for optimal paths. This theoretically will make the various races more predictable. In addition, it's a LOT of work for very little benefit to the playing experience.

The beauty of open-sourcing the game will mean that players who disagree with me will be able to easily change things more to their liking.

 

[RayF]

The Darlok Diplomat...

“Greetings, my good friend. We are perfectly normal living creatures just like you. Nothing mysterious here."
"Perhaps you would be interested in a trade treaty?”

 

[RayF]

The Darlok Spy

 

[kyrub]

I always wonder if people mean this opinion seriously. The other possibility is more probable: I am an obsolete gamer. Everybody else just wants to be tricked and "entertained".

Taking this to the current extreme; would you be willing to play chess exclusively against the top difficulty of one of the modern engines? A game on equal terms relies on the opposing number actually doing mistakes for you to exploit.

Sorry? You must be living on another planet. I have not seen one game with engines close to medium difficulty chess AI. Can you cite one?
The truth is, the developpers don't even try to do it. The resources, the technics and CPU power are all there, they just don't invest in decent AI. I don't ask for any revolution, just for a decent AI opposition (with bonuses, why not).

(I feel stupid I have to explaing this on Kodex, of all places.)

 

[RayF]

Sorry? You must be living on another planet. I have not seen one game with engines close to medium difficulty chess AI. Can you cite one?
The truth is, the developpers don't even try to do it. The resources, the technics and CPU power are all there, they just don't invest in decent AI. I don't ask for any revolution, just for a decent AI opposition (with bonuses, why not).

You are correct. There are no 4x games with an AI tuned to a difficulty equivalent with even a medium level chess AI. There's no market incentive for it. I am hoping to make the AI at least play an interesting game against a player with no bonuses, but a truly difficult AI will need to come from open-source work.

 

[RayF]

It's not a 4X game but was wondering what you thought of Arcen's approach to AI design for AI War (RTS or RTWP-S). http://arcengames.com/designing-emergent-ai-part-1-an-introduction/

I'm not really sure how a componentized AI like that would work in a 4X game. Ship combat is the only area that it might make sense and then you are never dealing with more than 6 stacks.

 

[Infinitum]

Sorry? You must be living on another planet. I have not seen one game with engines close to medium difficulty chess AI. Can you cite one?
The truth is, the developpers don't even try to do it. The resources, the technics and CPU power are all there, they just don't invest in decent AI. I don't ask for any revolution, just for a decent AI opposition (with bonuses, why not).

Of course not kyrub; no current game is nowhere near as solved as chess. I would be interested as to your thoughts what constitutes good ai however. Simply being able to solve the game and finding the strongest/most reasonable move in itself ultimately isn't very interesting as once it underststand game concepts well enough to iterate you end up with the aforementioned chess siliness as say the top bots outranking the top human player by some 500 ELO points. Substitute chess with checkers or mandala or whatever for an even more lopsided example. So assuming you can indeed teach the computer to play your computer game of choice, the next step would invariably involve teaching it to make avoidable mistakes to make the game playable at lower difficulty levels (unless you use different algorithms or manually limit its ability to calculate I suppose). Giving you the illusion of a good ai if you will. I recall seeing a rather lengthy tl;dr somewhere else on the same subject on the codex?

Of course, given the rng and random world generation the creation of that type of an ai would probably not be feasible for computer games, but given the advances in machine learning (AlphaGo and whatnot) we might see something similar should the economic incitament exist (or once the technology gets commonplace enough for monocled modders such as yourself). Again though, why would anyone want to play seriously against a genuinely difficult ai (ie one that can play the game and outclass human opponents in calculating possible scenarios) as opposed to one that makes mistakes (reasonable or otherwise) by omission or design?

Also yes I know we don't talk about the same thing but as you mentioned we're on the goddamn codex. What is an RPG AI? Humor me plx.

 

[RayF]

This is the direction of the Darlok Council screen

 

[RayF]

You can see the improve UI in action here:

 

[RayF]

Here's a "State of the Game" update for those interested in project details: https://remnantsoftheprecursors.com/2016/11/12/state-of-the-game/

 

[RayF]

Guys, I'm making the development builds publicly available through December since it's a long way to the next alpha. Should be a good opportunity for anyoone who wants to see the current state of the game.

https://remnantsoftheprecursors.com/2016/12/02/its-development-build-december/

 

[RayF]

hai guys.

Here is some almost-done artwork for the GNN robot for the Darloks.

 

[spectre]

The image looks great, but the animation feels a bit off. The rings at the darlok's fingertips are the main culprit - they look mis-aligned with the hexagonal grid resulting in a weird perspective - as if they were on a different plane.
Two relatively simple fixes I can think of is make the guy's hands move in a straight line, along the X axis, or get rid of the rings altogether and make individual hexes on the grid light up and fade as they are touched.

 

[thesheeep]

Gotta agree there, the finger movement looks disconnected from the hex grid they are on.

 

[Lagi]

Look interesting
can I somewhere try the demo/alpha ?

Darklok fingers are fine.

 

[Luka-boy]

It's just for testing purposes but holy shit

 

[RayF]

So, the real value in creating gigantic galaxies is to find inefficiencies in the way data is created and handled. Attempting 50K stars quickly exposed two different performance issues that were making the galaxy generation take too long. Figuring out how to offload that greatly improves start time (which was already fast compared to similar games). In addition, one of these issues was also occurring in the "Next Turn" logic, so fixing it also improved Next Turn processing times. This difference was especially noticeable on larger maps.

This is why stress testing is an important part of software development. I'm pretty sure that most AAA companies only make a token gesture towards this nowadays -- it's much easier just to tell gamers they need a beefier PC.

If you think 50K is a lot, I should point out that I am currently working on a major refactoring of the data structures to allow much, much larger galaxies to be generated in the same amount of memory. If I had to guess, I'd say that a galaxy of a half-million stars will be possible on my 16G PC -- 10x what is in that video. That would translate to a generated galaxy of about 5000 light-years across.

No one would ever play in something like that, but what it really means is that someone with a Raspberry Pi (1G of memory), might be able to easily play in a 1000-star galaxy that games like Paradox or nuMOO might require 16G or more.