Forums » Suggestions
Increase the number of Stygian Furie allowed to be launched at once
Currently the max number of furies allowed to be manufactured and launched at one time is three. When the Furie was first introduced, this number may have been sufficient. With the addition of powerful capital defensive weapons, such as the capital rail cannon (that does 10k damage per shot), this number is no longer sufficient.
Allow us to manu and launch nine or more furies.
Allow us to manu and launch nine or more furies.
+1. Considering the fact that furies kill themselves or get lost half way to where your going this might actually mean you can get 3 furies to the destination finally. Not to mention this is easily counterable by capship TU and cap swarm.
Considering the fact that furies kill themselves or get lost half way to where your going this might actually mean you can get 3 furies to the destination finally.
Is this actually a current fact, or are you referencing problems with behaviour from the past? A lot of things have changed in the last couple of months.
Is this actually a current fact, or are you referencing problems with behaviour from the past? A lot of things have changed in the last couple of months.
Current problem. If I need to I'll even go and make more just to test again if that helps
This is one reason why I asked if the AI for the Furies had been modified. They are/were kinda dumb and not very useful; hence my not using them.
If the AI has changed, then maybe they're "smarter" now?
If the AI has changed, then maybe they're "smarter" now?
If the AI has changed, then maybe they're "smarter" now?
Hence my question, since all AI has changed in the last month or two.
Specifically, things like bots getting "lost" should be getting pretty uncommon.
It is a progressive refinement thing, where, as bots are present in a sector more, they will establish more paths through that sector and be able to path-find more efficiently. This evolution is sector-by-sector, so some of them have better navigation paths than others, but bots getting truly stuck should be uncommon now.
However, if IronLord had a specific and reproducible case in which bots are being "lost", that should be eposted to Bugs (with a very carefully stated test-case).. but not used as a justification for a feature change.
Really, it's best to fix bugs, and then make the game that we want. Not hack some alternate game that's built around the bugs.
Hence my question, since all AI has changed in the last month or two.
Specifically, things like bots getting "lost" should be getting pretty uncommon.
It is a progressive refinement thing, where, as bots are present in a sector more, they will establish more paths through that sector and be able to path-find more efficiently. This evolution is sector-by-sector, so some of them have better navigation paths than others, but bots getting truly stuck should be uncommon now.
However, if IronLord had a specific and reproducible case in which bots are being "lost", that should be eposted to Bugs (with a very carefully stated test-case).. but not used as a justification for a feature change.
Really, it's best to fix bugs, and then make the game that we want. Not hack some alternate game that's built around the bugs.
What you say makes perfect sense, however this part is confusing: "It is a progressive refinement thing, where, as bots are present in a sector more, they will establish more paths through that sector and be able to path-find more efficiently"...
Does this mean that all bots share a common memory/map for path finding? So, in theory, the Furies learn from the hive's path-finding activities? As would the Unrats, convoys, SF, etc...
If I have time, I'll go play with the furies a bit, but something tells me that the hive hasn't had an opportunity to really map out common use (low hive activity) routes.
Does this mean that all bots share a common memory/map for path finding? So, in theory, the Furies learn from the hive's path-finding activities? As would the Unrats, convoys, SF, etc...
If I have time, I'll go play with the furies a bit, but something tells me that the hive hasn't had an opportunity to really map out common use (low hive activity) routes.
So first of all, just to be clear, I don't know that this AI stuff has anything to do with why Furies were being "lost". I would need more information on what was happening.
For instance, we had problems with NPC convoys awhile back, which had very little to do with AI, and was more of a data-handling problem between sectors, and the way things were timed. So.. AI may not bear on this issue at all. That's why we would need a specific Bugs post, with a reproduction case, to investigate.
Does this mean that all bots share a common memory/map for path finding?
Yes. Basically, we use our own heavily-altered and customized version of what is termed probabilistic roadmaps (PRMs). Our implementation actually diverges quite a bit from the original paper, or how most other people seem to use it, but basically there are various "nodes" (navigable points in space) and "paths" (routes) between the nodes.
Then there's also a route-optimization algorithm that helps refine the specific route to the goals of the specific NPC, and also a lot of specialized rules that check whether they can "directly" see their goal, or a given node, etc.
There are two unrelated sets of nodes and paths, one for smaller ships, measured more-or-less at the radius of the Behemoth, and another that is specific to capships. All NPCs of each "class" share the same PRM data.
We have an algorithm that basically continually refines the paths and nodes in the given sector, based on the success (or failure) of the NPCs flying through it, and what they find. So, essentially, the NPCs "explore" the sector, and are used to discover potential locations for new nodes, and to connect the nodes together.
It would be pretty time consuming, and computationally expensive, to come up with near-optimal solutions for all ~3000 secotors that have asteroid content in them. Plus, this data would be invalidated if anything moved, and we're working towards a goal of letting the universe be a bit more dynamic and uncertain.
As a result, the system is currently built around the NPCs being able to "learn" any new sector and improve the navigability of that sector, over time. This is also useful to potential future expandability of the universe.
It should be noted that while PRMs are an optimal way that NPCs find their way around, it is not the only system in place. There's actually a whole collection of different algorithms that have various fallbacks and priorities, to allow the bots to adapt to scenarios where there isn't much PRM data, or where there are dynamic objects in the way (like a bunch of capships, for instance), or other situations.
Similarly, PRMs are only relevant if there's a dense enough sector to obscure the NPC's goal, or make it difficult to reach. Like Bractus P1, for instance. Most sectors in the game do not have this level of density, which is why the "old" (2019-and-earlier) standard of AI tuning was usually sufficient, in the past.
We spent a lot of time testing, tuning and improving this in Bractus P1 over January and February, as well as another sector in dev-space, to prevent NPCs from getting stuck in local minima, seeing how quickly they can react to different conditions, and the like. We also used Dau L10, Odia M14, Itan J11, Sol H13 and the "training sector" (initial newbie zones) as test-cases for how quickly and effectively the PRMs were being generated, and how well the bots of various sizes were handling the conditions.
For instance, we had problems with NPC convoys awhile back, which had very little to do with AI, and was more of a data-handling problem between sectors, and the way things were timed. So.. AI may not bear on this issue at all. That's why we would need a specific Bugs post, with a reproduction case, to investigate.
Does this mean that all bots share a common memory/map for path finding?
Yes. Basically, we use our own heavily-altered and customized version of what is termed probabilistic roadmaps (PRMs). Our implementation actually diverges quite a bit from the original paper, or how most other people seem to use it, but basically there are various "nodes" (navigable points in space) and "paths" (routes) between the nodes.
Then there's also a route-optimization algorithm that helps refine the specific route to the goals of the specific NPC, and also a lot of specialized rules that check whether they can "directly" see their goal, or a given node, etc.
There are two unrelated sets of nodes and paths, one for smaller ships, measured more-or-less at the radius of the Behemoth, and another that is specific to capships. All NPCs of each "class" share the same PRM data.
We have an algorithm that basically continually refines the paths and nodes in the given sector, based on the success (or failure) of the NPCs flying through it, and what they find. So, essentially, the NPCs "explore" the sector, and are used to discover potential locations for new nodes, and to connect the nodes together.
It would be pretty time consuming, and computationally expensive, to come up with near-optimal solutions for all ~3000 secotors that have asteroid content in them. Plus, this data would be invalidated if anything moved, and we're working towards a goal of letting the universe be a bit more dynamic and uncertain.
As a result, the system is currently built around the NPCs being able to "learn" any new sector and improve the navigability of that sector, over time. This is also useful to potential future expandability of the universe.
It should be noted that while PRMs are an optimal way that NPCs find their way around, it is not the only system in place. There's actually a whole collection of different algorithms that have various fallbacks and priorities, to allow the bots to adapt to scenarios where there isn't much PRM data, or where there are dynamic objects in the way (like a bunch of capships, for instance), or other situations.
Similarly, PRMs are only relevant if there's a dense enough sector to obscure the NPC's goal, or make it difficult to reach. Like Bractus P1, for instance. Most sectors in the game do not have this level of density, which is why the "old" (2019-and-earlier) standard of AI tuning was usually sufficient, in the past.
We spent a lot of time testing, tuning and improving this in Bractus P1 over January and February, as well as another sector in dev-space, to prevent NPCs from getting stuck in local minima, seeing how quickly they can react to different conditions, and the like. We also used Dau L10, Odia M14, Itan J11, Sol H13 and the "training sector" (initial newbie zones) as test-cases for how quickly and effectively the PRMs were being generated, and how well the bots of various sizes were handling the conditions.
Inc, this is why I like you, man. Ask a yes/no question, and I get the answer to the question I was too afraid to waste your time by asking!
I'm fascinated by the level of intelligence you guys are trying to build into the game. Your application of Narrow AI to the game mechanics is definitely improving the game play.
As to this: So first of all, just to be clear, I don't know that this AI stuff has anything to do with why Furies were being "lost" -- yes, I took you on a bit of a tangent, apologies to the original thread creator. I will say that the Furies (possibly different now, I acknowledge) tended to get stuck. if you wanted to use them for some high-risk mining guards, or escorts, or security, they'd get hung up slamming into asteroids or lost when jumping, especially if you landed in a storm.
I have no idea if path finding has/had anything to do with that.
Thanks for your time!
and, +1 to increasing the number of Furies that can be launched at once; and I'd like to add a request to allow them to dock in my Goli and not have that count as flying time (suspend the countdown.)
I'm fascinated by the level of intelligence you guys are trying to build into the game. Your application of Narrow AI to the game mechanics is definitely improving the game play.
As to this: So first of all, just to be clear, I don't know that this AI stuff has anything to do with why Furies were being "lost" -- yes, I took you on a bit of a tangent, apologies to the original thread creator. I will say that the Furies (possibly different now, I acknowledge) tended to get stuck. if you wanted to use them for some high-risk mining guards, or escorts, or security, they'd get hung up slamming into asteroids or lost when jumping, especially if you landed in a storm.
I have no idea if path finding has/had anything to do with that.
Thanks for your time!
and, +1 to increasing the number of Furies that can be launched at once; and I'd like to add a request to allow them to dock in my Goli and not have that count as flying time (suspend the countdown.)