[TEC] Adding new sensors

[tater]

Is adding a new sensor as easy as cloning (copying one already there, and pasting it in as a new node, assuming that is "cloning") a sensor node in AI_Sensors.dat?

I can see the stuff in there with S3D, but I'm unclear on if it will work, and how to do it.

I had an idea which was to make a new visual sensor, AI_Visual_Naval, and that would replace AI_Visual on most warships. It would be set where AI_Visual is now, and the old AI_Visual would then be dropped to a lower sensitivity, range, etc. The difference would be the number of crew, and their training. One goal would be to make it so that if you managed to get past the escorts, the merchants would have a lesser ability to detect you, ideally making night surface attacks possible where they are not currently.

tater

[skwasjer]

Yes, you can. Export the nodes and reimport them back (use insert). Change the id's/parent id's so you have an exact copy (no clone function yet in S3D), but differently identified. You'll have to clone all related nodes too (in this case 4 total). Then, once you've got the copy, rename the label to AI_Visual_Naval. Then, edit the ships .sns file (with notepad) and modify the O01 node section.

Be advised, this is based on my insights, not on actual tests :P

[tater]

I might try this tonight.

tater

[tater]

Should I change the IDs, THEN export it (without saving), then open the original copy again and import?

I have to admit, this confuses me, lol

I can change the IDs on 2 nodes, but the stuff below belongs to them, and I only have a link, nothing I can edit.

[tater]

Ah ha!

export them.

change IDs in working version for new sensor.

import them, since they have old IDs, they nest

The trick is you need to switch from tree to list for it to show!

tater

[skwasjer]

With id's shown as links, use CTRL + click on link to switch to edit mode.

I know the export/import procedure is a bit wonky but it should work.

[tater]

Right now I've done some control tests to start. Single large merchant amd a SC in a line. I approach from the front quarter. I tested stock, and the stock sensors DAT combined with the 2 cfg files from TM1.6. All at 8 knots.

In both control cases, the detection is VERY sensitive to aspect ratio. If I keep my nose at the SC, I can get inside 500 yards before I get a reaction. I can steam parallel at 1nm with no problem in either., too. With the TM cfgs, I get detected showing more than the bow at 0.7nm. I haven't even tried the altered dat yet since I got so damn close in both stock and TM values.

I think I need to make a bigger convoy so I can't control what I present to the target so easily, this is very hard to test.

tater

[tater]

Convoy of 8 merchants in 2 columns, 800m apart, escorted by 4 Minekaze (all set to AI=3). The escorts were at 12, 3, 6, and 9 o'clock. Radar fix mod in place to assure visual not radar sighting.

Control (TM cfgs), stock dat:

DD detected me ~2000 yards at 8 knots.

Modded dat (AI sensitivity set to 0.03, range set to 6000 max), TM cfgs:

DD detected me (quartering aspect) at 1400 @8knots.

Note that I didn;t set the DDs to have better AI, everyone was using the dumbed down version. So if the mod was done right now, the warships would have detected me at 2000, and the merchants at 1400 in the last example. This was in the default weather conditions, so pretty easy viewing.

I think I coudl have gotten closer to the DD had I given him less to look at, but at some point I'd have to show him a broadside.

You'd have to play this many times to get a feel. I noted with the dumbed down sensors, the DDs kept their lights on when I was in the convoy, but didn't shoot as much as I'd expect. Another test would be the same convoy minus the DDs, then also try various AI levels and see how that moderates things. If AI skill has any bearing at all, I bet you can get withing spitting distance of "Novice" merchants.

[tater]

I'll make a quick mod with 2 sensors, AI_Visual, and AI_Visual_Naval insde the dat, and I'll throw in a DD sns with Naval set, and maybe a merchant with Naval set. I can make a mission with the "Naval" merchie convoy north, and a dumbed down visual merchie convoy south. That will simplify testing because it'll all be in 1 mission.I can even have it generat a convoy every 2 hours or something, so we can test a few times with 1 mission load.

Then we can test different AOBs, etc, and see how the sensitivity affects detection on the good vs crappy AI settings. More eyeballs on this are better, and I have other stuff I need to do, lol.

If it appears to be about right, then a 2d mission can be in there with the same convoys, but with escorts set to "Naval."

[tater]

Tested a convoy (no escorts) with "naval" visual sensors (current stock setting renamed), and another with the sensitivity dropped to 0.03 and range 600 max.

The naval (stock) I got detected at 1500 yards (EDITED, i remembered wrong) , and the dumbed down version was under 1000 yards. Both were about 45 degrees AOB and 8 knots using the TM cfg settings.

[tater]

http://mpgtext.net/subshare/434AI_sensors%20mod.rar

ALPHA test mod. This is the first time I ever cloned anything, or inserted a node in a dat. Might blow everything up, lol. Entirely not ready for prime time.

Makes 2 sensors, the default AI_Visual becomes dumbed down, and current stock becomes "Naval." There are 2 cfg files taken from TM in there, too.

The mission is 2 convoys steaming west, one North, one South. North is "Naval" visual (stock), South is dumbed down. No escorts, AI=competent. If you want to make a test mission, I included a minekaze with "naval" visuals and no other sensors, and a Mutsuki with stock sensors (dumbed down with the mod), and also no other sensors. The Nagara sns in there is to eliminate the hydrophone, I wanted the test to be purely visual.

Note that this only works as a test since to work generally, I need to alter every single warship sns in the game. I suppose doing it the other way might make more sense since there are more warships than merchants...

[skwasjer]

I can't test for you because I'm working on ehm well, you know, but I'm glad you got a bit further using my tip

[tater]

I intend to update this test mod in the somewhat near future.

A few questions for anyone who might know:

1. will the game recognize 2 of the same general sensor type installed? I tried cloning the O node on a DD, and renaming it O02 to add a 2d visual sensor, but it didn't seem to work. It could very well be a failure in my ability to properly work the nodes, however. The goal here would be 2 concentric AI_visuals on 1 ship. One longer ranged version, and one very short ranged version. In my tests I didn't have them overlap, so one goes from 2000m to max rannge, the other goes from 0 to 1999m.

2. Any insights into proper settings for AI visuals to detect a certain cross sectional area at a given range X% of the time?

3. Does "alert" status for AI ships alter their detection probabilities at all?

4. The default game has the "use crew skill" (whatever it's called in sensors.cfg) set to false. What's the effect with this on?

Musings about the requirements for such a mod in terms of realism:

Strictly speaking, with no consideration but a detector at the highest point on the ship, detections ranges should be 100% reciprocal. I see your eyeball, you see mine. That said, the eyeball in space at the tip of the mast is not reality. This boundary case does apply to first detection, however.

The detector (the watch crew) are not at the very top of the ship. So all things being equal, the ship with the largest cross sectional area above the watch crews gets detected first. I am in the shears with 2m^2 of metal above my eyeballs, and the Fuso I'm observing has 20m^2 of pagoda above the observer there. I suppose we could get into typical resolution limits for the naked eye, or even normal telescopes (those without adaptive optics) and figure out what minimal cross section should be invisible at a given range—actually, that might be a useful calculation for a boundary condition... (it'll give me a chance to brush up, I'd almost forgotten I was an astronomer in training once )

Luckily, for submarines this is somewhat straightforward, anyway. The guys in the shears are still close to the water, and the cross sectional areas involved are tiny in the case of a sub compared to a ship. The short answer is that a submarine should, barring a human failure, ALWAYS visually detect a ship first. This is particularly true if you consider smoke, which functions as a large, detectable cross section (even for pretty non-smokey ships) that resides well above the ship's watch crew.

Seriously, I cannot imagine a realistic case where this is not true, with the only exception being crew skill. Put a myopic guy without glasses in the shears, and all bets are off (assuming tojo in the BB's mast isn't the coke bottle wearing ww2 stereotype), but assuming equal AI skills, the ship should always be seen first IMO. The game limitation of visual limit IS an issue here, because even set to 16km, the sub would have detected the ships outside the visual limit (capital ships, anyway). At 7-8km, we are in a situation where the targets may very well have already detected the sub in RL, but the sub should still be aware of them first.

Once inside the extreme range case (top of masts or smoke on the horizon), things change a little. Once both ships' watch crews are well above the horizon to each other, you have to take the cross sectional area above the horizon (to the other observer) into account. So the next obvious test case would be a range at which the sub is entirely hull up.

It's still a small target, so the chances of any given set of eyeballs on the ship looking for it should be lower than the chances for the sub to see the part of the ship above the horizon. The ship "wins" in terms of simple geometry (higher mast), but the cross section of the sub is smaller than the hull-up portion of the ship. The difference happens when you look at watch crew numbers, and their eqipment. At a certain point, the ships start having more watch crew. So the smaller instantaneous chances of detecting the sub are mitigated by more eyeballs. Better optical gear could have an effect as well.

In general, I'd expect the sub to always spot ships visually first at extreme range (clear days). There are zero plausible exceptions to this, IMO.

Once the sub is hull up to the ships, it becomes more complex. The ships should always be visible to the sub, obviously, but their chance of seeing him should increase. As more of the ship is hull up, more eyeballs come to bear. It might be interesting to know how many watch crew a typical IJN DD stood vs a CA, BB, etc. Comparing that to a sub's watch would give a relative number of eyeballs, and that would scale in some fashion to detection.

Right now internally I have 4 AI visuals (or 5, depending on which version I use, lol). I might increase this a bit.

My thought was to set them primarily based upon the expected number of watch crew, which would also scale to ship size/height (all things bein equal, the higher the watch, the better).

IJN BBs had watches up as high as 150 ft, though perhaps 100-120 was where the bulk of them would be. CAs perhaps 60-75'. DDs have what looks like a crow's nest in the mainmast at ~60', but that would be only a few of the watch crew, the bulk look to be more like 30', tops. Subs look to have a watch someplace in a similar range up in the shears, 20-30'. Merchants have some height to work with, but in most cases they'd simply not have the large number of trained eyeballs as a warship.

So I'm thinking (some should be combined if they are similar in effect):

1. Merchant (few watch crew (30 - 45' max watch, few crew))
2. Merchant 2 (more watch crew (45' - 60' max watch))
3. Small warships (subchasers (45' max watch))
4. DD (60' max watch)
5. CL/CA (CVs would be here, too (75' max watch))
6. BB (150' max watch)
7. Single crew aircraft
8. multi crew aircraft

They are arranged in order of detection probability/distance. All but the aircraft should have the sub detecting the ship first, no exceptions. The planes are smaller, and on top of that occupy a large volume of space instead of a linear position on the horizon. Getting jumped by a plane if you don't have radar should be likely, getting jumped by a ship in clear weather should be completely impossible.

2 and 3 should probably be combined. I might use that level only on certain merchants of high value. That would make 7 visuals. A simple way to create variability for warships would be to "BP-clone" warships to make an additional class for a few. The class would get a ship name within the real class. The clone would be identical except for having the visual set 1 notch lower. Then you might come across 2 asashios with std DD visuals, and 1 Kagero, Isokage, whatever, with "small warship" visuals (crew had too much sake on liberty )

The goal would be accurate outcomes, not strictly accurate visual limitations since there are simply not enough variables to work with set them to what we know a RL limit is, and they might be superhuman in game in terms of spotting.

To start I will assume the stock view distance, for example. The benchmark is sub detection for the AI, and ship detection for the sub. The sub should always see ships first, and the bigger ships should pick up the sub faster than the smaller ones all things being equal. We know from patrol reports that it should be possible to conduct night surafce attacks, and even that some subs got within under 1000 yards of escorts undetected. This should be possible with good sub handling, but it should not be automatic, IMO.

I'll add a couple more to my test.

Note when I release this as a test, I will likely simply make a set of ships with 1 of each sensor type, and a test mission for you. To control the test, I will eliminate all other sensors from the test ships so we can get decent results.

[Rockin Robbins]

Originally Posted by tater

http://mpgtext.net/subshare/434AI_sensors%20mod.rar

ALPHA test mod. This is the first time I ever cloned anything, or inserted a node in a dat. Might blow everything up, lol. Entirely not ready for prime time.

Makes 2 sensors, the default AI_Visual becomes dumbed down, and current stock becomes "Naval." There are 2 cfg files taken from TM in there, too.

The mission is 2 convoys steaming west, one North, one South. North is "Naval" visual (stock), South is dumbed down. No escorts, AI=competent. If you want to make a test mission, I included a minekaze with "naval" visuals and no other sensors, and a Mutsuki with stock sensors (dumbed down with the mod), and also no other sensors. The Nagara sns in there is to eliminate the hydrophone, I wanted the test to be purely visual.

Note that this only works as a test since to work generally, I need to alter every single warship sns in the game. I suppose doing it the other way might make more sense since there are more warships than merchants...

Tater, I'm sure you realize how BIG this is. It's like being in the room when Alexander Graham Bell says "Watson, come here, I need you." Ok, maybe not quite that big... lol

[tater]

It's already been done, actually, that's my understanding. I'm just using a similar notion for SH4.

There are other novel approaches to visual detection possible as well, but they need to meet certain standards to be realistic.

Goals:

Sub should always detect ships first assuming distant visual range, ideally. Ship size vs sub size, plus smoke ensure this is the case.

Given the short visual limit compared to RL (the view distance well below the max possible range to see a ship), there are some issues to be addressed. A sub could very well be detected by a ship beyond what the player can see in game. OTOH, the sub should know about it first. That's really the problem. If a TF should have spotted the sub on the surface at 15km and the vis range is 8km, the DDs should ideally be heading your way with a bone in their teeth. OTOH, you should know this before the shells are raining down since you'd have been watching them charge in for 7km. Unfortunately even a 16km view distance is not far enough. Bleeding edge limit for an IJN DD is likely ~22km based on geometry to be able to see the hull up conning tower of a sub. 14-18km is more likely. That's not a certain detection, just possible. For larger ships it's farther, it in fact pushes the render distance where ships are no longer abstracted (20nm or so I believe).

In game this means relying on the watch crew. They'll call out a target bearing 237, and you will look and not see it.

tater