Gunnery. Big guns, secondaries, speed, evasion and so forth..

[akkarfjes]

Let me first say that I love what has been done so far, and I am looking forward to what we have coming. I have been playing RTW/RTW2 a lot, and I am looking forward to a different approach to the subject and era.  Now to my point:

I am not sure if this has been brought up before, but to me it seems that gunnery in UA:D (and RTW for that matter), is done backwards. Let me explain: In UA:D you acquire a target. Aim your guns at that target and fire at said target. Whether you hit or not is determined by a random roll influenced by several variables. Would it not be easier to acquire target, estimate a point in space where the target will be in a point in time, and try to hit that point. 

Many of the variables will be the same, but as they will become more integral in the modelling they will make more sense than any percentage value set to hit. Speed will influence aiming only if change in bearing to target exceeds your traverse speed.  Any change in the targets vector will affect the chance to hit if  it brings the target further away from the target area at the time of impact. 

I think that aiming at a point in is a superior method in many aspects, mainly:

• More realistic
• You will have two separate processes, each with fewer variables (Where shells land, and where ships are)
• You will make for a more realistic modelling of secondaries vs Main guns as small calibers guns will have less time in flight, and therefore a lesser Delta for the targets position.
• You make it possible to hit other targets than intended.
• You will make the speed/evasion discussion moot.

 

[Cabusha]

Definitely agreed.  The RNG hit percentages are pretty annoying and abusable.

[Hangar18]

21 hours ago, Cabusha said:

Definitely agreed.  The RNG hit percentages are pretty annoying and abusable.

It's impossible to make a gunnery model without rng...

[coalminer]

On 8/3/2020 at 12:27 PM, Hangar18 said:

It's impossible to make a gunnery model without rng...

I think its more to have/allow the player to manually lead targets by setting where the player thinks the enemy ship would be and the AI crew calculate the firing solutions to hit that X target than to fully remove RNG from the calculations. Although this does make the game a little skewed towards either the player/AI when technology differences come in as manual leading can result in "easy sniping" of the AI with very large calibre guns of decent accuracy as the AI may be using a different (existing) system of gunnery.

Although this does help in certain issues like the AI crew randomly deciding to change targets and shells coming out (seemingly) completely skewed in flight due to how accuracy is calculated relative to the enemy vessel than the position being aimed at. so RNG should be determined at the gun barrel and shell should not deviate too much down range, what matters more is the target lead and range being applied and less set the game to x5 and hope the gunners arent drunk.

Edited August 5, 2020 by coalminer

[akd]

That has very little to do with actual gunnery.  You don’t “know” definitively where the target is in relation to your estimate of future target position.  Showing an aiming point in relation to an actual position is a gross simplification of the gunnery problem that removes a number of important variables (rangefinder error and course error in particular).  An abstracted system could produce more plausible outcomes, although the current abstracted system does not.

[Staire]

I agree entirely although I think it highly unlikely they will move in this direction now seeing the size of the dev team + how much they have already committed to their current model.

[DeadlyWalrus]

Yeah, I kind of feel that completely reworking the gunnery model might be a bit too much for them to pull off.  Not only would they need to rebuild the gunnery system itself, but the whole ship handling AI would need to get much more detailed and capable than it currently is.  Individually maneuvering ships that are under fire is all well and good in one of the small scenario battles we have right now, but having to do so in a major fleet engagement would quickly devolve into micromanagement hell.  While properly simulating all of this would be quite nice and would eliminate the need to manually balance a lot of the gunnery variables, I'm just not sure that it's a realistic proposition.

[Hangar18]

On 8/4/2020 at 11:12 PM, coalminer said:

I think its more to have/allow the player to manually lead targets by setting where the player thinks the enemy ship would be and the AI crew calculate the firing solutions to hit that X target than to fully remove RNG from the calculations. Although this does make the game a little skewed towards either the player/AI when technology differences come in as manual leading can result in "easy sniping" of the AI with very large calibre guns of decent accuracy as the AI may be using a different (existing) system of gunnery.

Although this does help in certain issues like the AI crew randomly deciding to change targets and shells coming out (seemingly) completely skewed in flight due to how accuracy is calculated relative to the enemy vessel than the position being aimed at. so RNG should be determined at the gun barrel and shell should not deviate too much down range, what matters more is the target lead and range being applied and less set the game to x5 and hope the gunners arent drunk.

Ummmm no...the player is supposed to be a commander, not god.

[Entropy Avatar]

I am definitely more in favour of a simulationist model of gunnery. Correctly estimating the hit chance under all possible conditions (and how all possible conditions affect hit chance) is a very difficult problem. Deciding what the hit chance should be and then forcing shells to hit or miss can look super bad in some conditions. Having a system where you try to estimate where to aim and then model where the shells falls just seems easier and less error-prone. You can break down each component and think of it in real terms. E.g. this range-finder tech should have this much uncertainty at this range. This estimate of ship speed has this uncertainty function. 

I guess one drawback of the simulation approach is it may be harder to explain to the play why he is missing without a "real" (forced) probability of hit breakdown. However, I think presenting uncertainty in target range, heading and speed (and observed hit rate) would feel a lot more immersive.

 

Edited August 7, 2020 by Entropy Avatar

[Cptbarney]

6 minutes ago, Entropy Avatar said:

I am definitely more in favour of a simulationist model of gunnery. Correctly estimating the hit chance under all possible conditions (and how all possible conditions affect hit chance) is a very difficult problem. Deciding what the hit chance should be and then forcing shells to hit or miss can look super bad in some conditions. Having a system where you try to estimate where to aim and then model where the shells falls just seems easier and less error-prone. You can break down each component and think of it in real terms. E.g. this range-finder tech should have this much uncertainty at this range. This estimate of ship speed has this uncertainty function. 

I guess one drawback of the simulation approach is it may be harder to explain to the play why he is missing without a "real" (forced) probability of hit breakdown. However, I think presenting uncertainty in target range, heading and speed (and observed hit rate) would feel a lot more immersive.

 

Im guessing they will add features to hide various stats for ships so you don't have some smoll version of omnipotence, also the armour model needs updateing/reworking since its too basic atm and causes a lot of zombie ship problems, combined with bow-tanking and abusing speed AI ships can get pretty hard to kill half the time, although i guess thankfully there is no reverse option yet otherwise that would become even more apprent.

[Sternack]

Looks like Atlantic Fleet uses point aiming with random dispersal. Only problem in Atlantic Fleet is that you have to more or less guess the point of aim, while historical real world fire control was 'computer aided' all the way back to 1902 with the invention of the Dumaresq. Dreadnought gunnery was not skeet shooting: it was an industrial process involving intricate time/decision loops, with many large sources of error. For example correcting off the wrong salvo in the time sequence generated a cascade of errors that took several minutes to recover from. Ultimate Admiral Dreadnoughts nicely models this common error by randomly losing target lock even when not maneuvering.

Prior to 1912 guns were individually aimed at the turret or gun, fired on the salvo bell and many shots did go wild due to sights vibrating loose, operator error, etc. UAD models this beautifully!

Director firing would be better modelled visually by keeping salvos falling in the same pattern but randomly vary the pattern width (dispersal) and point of aim, and only tie the fall of shot sprites to the salvo pattern during the last 1/2 second or so. I think Ultimate Admiral Dreadnoughts is trying to trace the entire path of the shell from gun muzzle to target. Not necessary to do this.

Many observers on the receiving end of dreadnought salvoes reported that ever shell in the salvo appeared to be headed directly for them, personally  - and it seemed a miraculous deliverance when the salvo missed. Programming this effect would be much easier than what UAD is trying to do and would look much more true to life (as it were).

[InsaneOne]

Going with the simulation group, I think it would be nice if aiming was done by a series of errors off of a perfect lead point, and done in steps. So essentially margins of error that cumulate along the steps of rangefinding, speed and direction finding, gun laying (with turrets sharing the same horizontal error across the guns in them), then timing with pitch and roll/stability (I think this would also affect RoF), with finally the spread of the gun itself. And then have the errors for range and lead be reduced with identification and shot correction, this would be hard to display as a % to hit though.

Some of the penalties to accuracy that are already applied would naturally still exist in this model, range errors are far more detrimental when shooting at small targets and speed estimation errors cause more issues with fast ships

[Jatzi]

On 8/8/2020 at 2:02 PM, Sternack said:

Looks like Atlantic Fleet uses point aiming with random dispersal. Only problem in Atlantic Fleet is that you have to more or less guess the point of aim, while historical real world fire control was 'computer aided' all the way back to 1902 with the invention of the Dumaresq. Dreadnought gunnery was not skeet shooting: it was an industrial process involving intricate time/decision loops, with many large sources of error. For example correcting off the wrong salvo in the time sequence generated a cascade of errors that took several minutes to recover from. Ultimate Admiral Dreadnoughts nicely models this common error by randomly losing target lock even when not maneuvering.

Prior to 1912 guns were individually aimed at the turret or gun, fired on the salvo bell and many shots did go wild due to sights vibrating loose, operator error, etc. UAD models this beautifully!

Director firing would be better modelled visually by keeping salvos falling in the same pattern but randomly vary the pattern width (dispersal) and point of aim, and only tie the fall of shot sprites to the salvo pattern during the last 1/2 second or so. I think Ultimate Admiral Dreadnoughts is trying to trace the entire path of the shell from gun muzzle to target. Not necessary to do this.

Many observers on the receiving end of dreadnought salvoes reported that ever shell in the salvo appeared to be headed directly for them, personally  - and it seemed a miraculous deliverance when the salvo missed. Programming this effect would be much easier than what UAD is trying to do and would look much more true to life (as it were).

Having watched some of Tortuga's Atlantic Fleet videos you don't have to guess. He reliably either hit or straddled his targets after only 1 or 2 ranging salvoes. I'm sure the system is fairly simple in Atlantic Fleet though.

[Entropy Avatar]

20 hours ago, InsaneOne said:

Going with the simulation group, I think it would be nice if aiming was done by a series of errors off of a perfect lead point, and done in steps. So essentially margins of error that cumulate along the steps of rangefinding, speed and direction finding, gun laying (with turrets sharing the same horizontal error across the guns in them), then timing with pitch and roll/stability (I think this would also affect RoF), with finally the spread of the gun itself. And then have the errors for range and lead be reduced with identification and shot correction, this would be hard to display as a % to hit though.

Yeah I think the current design is driven by wanting to have a hit percentage and breakdown of the terms in that hit percentage function. It's nice feedback to the player and can maybe explain to them why they are missing. But that's not something people really had. They had their methods and technology and that was that. What I like about the simulation approach is that it's very natural and problems at any stage will have natural consequences. So you don't have weird edge cases like "our bonus for close range isn't enough in situation X to counteract the penalty for small target size and high target speed" ergo we can't hit that DD that's 800 yards away and effectively stationary relative to us. In a "to hit" model, that can result from the function having slightly imbalanced terms.

[Shaftoe]

RNG aiming can be extremely frustrating, particulaarly to those whom RNGesus hates for no reason.

[Entropy Avatar]

4 hours ago, Shaftoe said:

RNG aiming can be extremely frustrating, particulaarly to those whom RNGesus hates for no reason.

To be clear any model of gunnery is going to have some random factors. In reality, there is going to be some "random noise" in each part of the process. E.g. a rangfinder is going to have some error with some standard deviation. When you estimate a targets speed, you will be off by some random amount, etc. So I'm not against aiming essentially being a function with some randomness and luck. If you've got a 3% hit chance, you might go a whole lot of salvos before getting a single hit (that then bounces...). Or you might penetrate that turrent roof and set off an ammo explosion on your very first full salvo. Them's the breaks!

 

I'm against how the function is currently designed though. Right now, every possible factor is boiled down to it's impact on hit chance. So in the end you have a bunch of deterministic calculations each affecting hit chance, combine them together and roll the dice to see if you hit or not. I'd much rather be rolling lots of dice, simulating the contributions of small errors in the aiming process at each stage. So instead of whoops, you missed, better luck with the RNG, it's more like "You missed because you under-estimated the target's speed by 2 knots, or over-estimated the distance by 2%. Going to adjust and try to dial it in.."