Serious Concerns About Skeleton Ships if AI is Not Improved in SoT

@williamherschel The better the AI you want, the more of a performance hit you will take. You can't get around it. One of the perks of having player run ships instead of AI ships is that the processing for behavior is offloaded from computational resources to our brains. And our brains are quite complex processors that are exceptionally difficult to duplicate.

Another thing is that it is really inefficient to brute force AI rules. You'd have to write a million different rules (Hyperbole, I know) to account for all the different conditions the AI will face from non-AI entities. Hence state checks and decision trees or machine learning if you need a very robust AI. Decision trees are nice since traversal time is limited to the length of the longest decision tree. Ideally you would turn it into a binary search tree or something like that for decent performance and speed. All you need is a series of boolean checks (hyper quick computations) with a completion flag. But of course you have to traverse the tree and that takes time and resources and trees can get quite large to account for large variety of states.

Now what could happen is that the AI is managed on its own thread completely server side. That way only the position/animation of each skeleton/entity has to be pushed to the client. Processing can happen between clock cycles and then pushes the new states to the client each cycle. Optimization can help with that since you can stuff more processing per cycle. Of course this could lead to instability if one thread hangs while the other is waiting for a response from it not unlike some communication issues from days of yore.

What I think would be really neat is if they implemented swarm rules. Instead of processing each and every entity by itself, you designate some entity(ies) as command entities and the rest as response entities. The response entities are computed based on the state of the command entity. If the command entity gets destroyed, you 'promote' a single response entity as the new commander and re-assign each responder to the new commander. To improve computation time, you have the tougher skeletons prioritized as commanders and weaker skeletons as responders. Since the tougher ones take longer to kill ideally, you promote less often. This could give rise to a variety of skeletons that follow a 'chain of command' if you will not unlike real life.

Just my two pieces of gold.

@williamherschel Yes Rare is the only one who can definitively say what is going on for the current AI implementation. I won't dispute that. Same goes with the existing AI. I wonder if Rare has put all of that on the backburner, including optimization, simply because of all of the demands for content. I sincerely hope that isn't the case here but it might explain the decreasing performance as assets and entities are added to the game.

Now in terms of what I said, I wasn't saying that the game specifically will take a performance hit for more complex AI behavior. Anytime you add things that require additional computational resources there is more demand on existing resources. That is something inherent in any kind of development and not specific to Sea of Thieves.

So to get around the increased computation demand, you have to take shortcuts or optimize. Swarm AI is an example of that. You only have one complex AI and the rest follow a simpler ruleset as compared to each AI being of increased complexity. You get more dynamic behavior and you reduce the added computational load. Note that adding such an AI may not actually be the best solution but it is definitely a solution.

Without optimizing the existing game, you have to take resources from something else to accommodate increased computational load. Right now it appears as though most of that comes from draw distance. Instead of computing where everything is so far out, you compute less and use those resources elsewhere. At best it is a temporary solution. I hope it doesn't become a permanent one.

But yeah, AI does need some work. Personally I would rather see the game get optimized more than have content added. Optimization would increase graphics performance and free up resources for more fancy things or future content.

In a recent UE4 dev talk about the shark AI a Rare employee said that all AI is handled server side and that the servers run on low power single core cloud servers that manage all the crews on a server and the waves and the AI.
Given how poorly the servers currently perform under load I don't see any possibility of better AI here.

If you want interesting and engaging combat Rare needs to flesh out the PvP aspect of the game.

@williamherschel said in Serious Concerns About Skeleton Ships if AI is Not Improved in SoT:

@program-024 said in Serious Concerns About Skeleton Ships if AI is Not Improved in SoT:

@williamherschel The better the AI you want, the more of a performance hit you will take.

I don't think anyone besides Rare devs can say this definitively. So while I also speculate this might be partially what's going on, it'd be nice to have some additional information about this issue. Especially when graphical performance (including draw distance) seems to decrease with every update. None of the AI threats in this game are fun to fight at all. I haven't met a single person who enjoys the Kraken or Meg encounter, and people are more forgiving of sharks and skeletons because they're essentially the one thing to do (PvE wise) in this game.

I just have a hard time believing that if any Rare tester or developer hopped into a higher level crew's game, and saw the level of kiting and persistent AI tracking issues on display, that they would say this is what they initially intended for this game.

The AI in SoT is bad, quite honestly some of the worst I've encountered in recent memory. I'm really not intending to be hyperbolic, but this is a major issue that the community continues to ignore in their continual persistence that this game needs more "content" above all else.

It's quite honestly a short sighted view, because that content will come, be easily dispatched, and then we will be back to square one where we have a bunch of AI threats in this game that quite honestly nobody wants to fight because they're simply not any fun.

Has anyone fought Meg post-HD? I'm guessing most have not. This is precisely the problem with SoT at the moment.

I wouldn't call the AI bad, its just that the AI can only be as good as what the environment and weapons they have.

Basically you have melee skellies and ranged skellies, the melee rush you and try to cut you down, while the ranged stay at their respective ranges firing at you.

Remember that there is no such thing as cover to speak off, no hiding behind low walls (ala Gears of War) no buildings to snipe from etc. etc.

Personally I think the AI is about as good as it needs to be, could it be improved upon? probably, but as someone else has pointed out, they are working on it. I mean we already now have skelly cannons that aren't top level sniper quality, and the same for ranged skellies that never seemed to miss.