Dev show 'n' tell stream feedback - extended horizon & airplane AI

Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#1
Sorry I couldn't make this stream, devs, it looked really great! Twitch VOD here or on YouTube:


To be clear, this was all *prototype* code - no implementation directly planned, yet. But it all looked so compelling, I'm really hoping that won't be too far off! (And also that console limitations don't block them from happening...) To summarise briefly what was shown:

@Django - coded and showed off how extending the terrain render distance would look under various conditions. How related aspects need tweaking to prevent players seeing the world tiles being loaded in (distance fog tweaks) and how illumination caching idiosyncrasies would currently make the world tiles far more visible, particularly under certain lighting conditions. Trees and other landscape objects were also continually popping into view at mid ranges.

Also, he had an optional change that made the horizon appear to be curved, as if it were a planet (as opposed to the ugly straight cut-off, currently). But this would apparently cause all manner of other code problems (e.g. with ray tracing not finding things where they appear to be, or something).

2019-10-10 Plane AI + Horizon.jpg

Then @Sdarks - Whipped out a squadron of enemy AI controlled planes which flew gracefully in formation, nose first into the ground. Lol! XD But then wowed us with a selection of different sized, randomly spawning planes (including one of mine:D) that did an very impressive job of flying about the sky in a natural looking fashion - just like other players in a co-op mode. None of the individual AI planes seemed to crash in the half hour demo (just the 'flocking' formations).

Apparently this code uses a PID (proportional integral derivative) control system that adapts to the plane's dynamic physical characteristics, setting it on the desired course. Pure player-type control inputs, no physics cheats! They aim to fly at the player, while staying around the same altitude (or just above when low to the ground).

I'll lay out my opinions and feedback below, along with other's viewers, hopefully.;)
 
Last edited:
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#2
(1) Increasing the render distance (to any extent) is great! It gives more location context, which is very handy in a game where many have often felt lost in similar repeating terrain, with few distinct landmarks. I could see 3 monuments simultaneously at some points in this demo, which would really help with this.

(2) The more distant horizon just looked so much better. Have you guys tried running the campaign intro with it like this? That often looks pretty bad, currently, particularly with the night-time draw distance during the day bug (is that glitch understood/fixed yet from, from looking at this @Django?). So extending the horizon could show a depth of scale to new players immediately on starting the game, here!:)

(3) In the first half of the stream, we saw all sorts of ugly visual artefacts with the lighting inadequacies, etc. But they didn't seem noticeable at all in the second half (with the AI planes). Is that because you were deliberately provoking them to start with? (Over-extending render distance? Moving time faster, etc?) I expected that looking out for enemy planes would make a great distraction to avoid noticing the issues as much, but I looked hard and couldn't see them anyway...

(4) The rounded earth effect looked so good... Are there really that many insurmountable technical issue with it? (Because, you know, things like reducing the processing time for converting old snapshots to the new system was sounding intractable, but then it just got sorted somehow... ;):D)

(5) And I'd say it doesn't necessarily contradict the infinite flat terrain of 'planet' TerraTech. I mean, you can just treat it as a camera effect - it's just like a fish-eye lens perspective, right?
 
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#3
(a) That plane control code looked an aweful lot more involved than just PID, as far as I know that technique (from it forming the core modules of my Cybernetics degree), @Sdarks! Presumably you have some kind of evaluation function to assess how well the AI plane is reaching it's desired trajectory. Or is that included in a code module or something?

Also, how are you even figuring out what trajectory to aim for? I mean, that's the big difficulty here, no? And how to roll and then pitch to be able to achieve these things seemed (to me, before hand) like it would be a very non-trivial AI issue....o_O


(b) I was kinda disappointed not to see, in the stream, how the planes fly around you when parked on the ground (looking up)... How well do they cope, currently? Pressumably they are totally blind to terrain height? So would tend to fly into tall mountains? How tricky might it be to have them pick themselves back up off the ground? e.g. in a height extended build beam?


(c) I've got properly flying techs in the game already, woot! :D (Not just stylised hovercraft.) The planes that you showed off were all extremely light weight, high wing ratio and very manoeuvrable. Did you test out heavier, chunkier (and far bigger) planes, too? e.g. little GSO "tube plane", longer more cumbersome "cluster drpship"...? (Others have better examples too, I'm not massively into planes.)


(d) Yours all had propellers, what about boosters (and hybrid fan/boosters)? Think it would be hard to code them to use that in a semi-intelligent, sparing manner?


(e) What about helicopters and anti-grav aircraft? I'd imagine that wing-less AG would be relatively straight forwards (with even simpler PID control). But throttled lift craft, yikes...o_O Maybe necessary to bypass the user-style interface of gradual throttle change? Then lean control for movement, etcetera I've no idea...

But if it 'learned' to fly planes...; I imagined that would *need* some kind of cheating system, but seeing the seeming miricle of these planes, I imagine you might be able to do the same again.;) Especially *if* you devs are going to be manually adding enemies, still - they can each be vetted to work, or at least automatically tested in simulation, or something...?


(f) I can see why Matt said these two features go so well together - I think you *need* extended render distance to give planes enough room to manoeuvrer. Let alone have them spawn in and a appear in a fairly natural way.


(g) It seems like enemy spawning would certainly have to be overhauled to coincide with (both of) these features. While it would be cool to see *some* planes appear while player is entirely on the ground, I think it best that they wouldn't ever (or very rarely) spontaneously agro onto ground techs. So like flying at 200m, passing by, etc. Maybe not have them appear at all until the player first flies - in the Venture mission or reaching a certain altitude in a tech with wings before that (maybe prompting a spontaneous new flight mission to pop...:)).

But also the de-spawn range would have to be significantly increased - looks like you had to do that for this demo, and waived the 200m immobilisation limit, too.

But with ground based enemies, too - I think this would mark time to make them less obviously spawn in and out around the player (breaking the immersion quite badly, currently). Being able to go attack an enemy one side without all the ones in the other direction magically vanishing would be handy. Also, perfect time to have enemy outposts (and clusters) spontaneously appearing in permanent positions.;)
 
Last edited:

Django

PAYLOAD STUDIOS
Joined
Apr 21, 2016
Messages
272
Likes
505
Points
505
Location
London, UK
#4
Thanks for the detailed feedback ZG. I'll try to provide some answers to the draw distance questions below

Have you guys tried running the campaign intro with it like this?
We have, and it really makes a difference! You can see far into the horizon and it makes you feel more part of the world imo.

Is that because you were deliberately provoking them to start with? (Over-extending render distance? Moving time faster, etc?)
I wasn't intentionally trying to provoke any, although accelerating the time may have had a slight effect on the tile lighting. Once dusk passed everything should have caught up which clearly it didn't. There are a few known changes that will improve this, and I have some ideas to try out.
Then there is the popping, and enemies at a distance in general.
Finally, the skybox and horizon also need a bit of work - the bright white in the day time is pretty glaring.
Are there really that many insurmountable technical issue with it?
Nothing horrible, but it comes with some new limitations. Bending the horizon will effectively limit our interaction distance proportional to how much we curvature we apply. At the displayed amount it would sit somewhere aroung 100-200 units. This is in the range of where we currently allow picking up blocks, but it would stop you from being able to interact with far away teleport towers.
 

RC-3197

Well-Known Member
Joined
Jan 5, 2018
Messages
23
Likes
51
Points
415
#5
I really love the features they were showing off in the stream, especially the Aircraft AI, I'm glad that a guy who worked on flight sims is working on that as nobody else in the world would do a better job than him.

The squadrons of airplane AI seem to be crashing in unison, is that because they try to fly before the PED system is activated? That's what it looks like to me.

I really like the idea of having squadron's of enemies, and if the AI gets advanced enough, wingmen following you around. It reminds me of one of my all time favorite games:

Star Wars Rogue Squadron

So basically, the airplane AI is extremely cool and I hope it is eventually in the game.

I'll also make sure to tune into the next livestream live next time.

However, at 26:44 they talked about crafting blocks in COOP:


The exact quote is:

"It would require a lot of work to add those into any sort of gamemode, coop gamemode not even on console, so we've had to put that on the backburner now."

The impression I get from this response is to not expect manufacturing blocks in coop any time soon, maybe not even a year or two from now. I hope I am wrong about this.

I get it, programming online play is notorious for being an extremely painful experience. At the same time, some things are worth the effort. In my personal opinion, crafting is just as essential to terratech as exploring or battling.

I've been doing COOP campaign with my brother (who introduced me to the game) for a few days now. We both are very experienced terratech players, we enjoy this game a ton, especially single player campaign. But we are definitely not enjoying COOP campaign nearly as much as single player campaign. I constantly find myself wanting to refine, craft, automine, and scrap. But alas we cannot. We must instead scrape up the scraps that the game gives us and hope we get what we want. There are some techs that I have accepted that I will never get in COOP campaign simply because of how expensive / rare the parts are, even though in the regular campaign I could craft everything I needed and pull it out. Crafting gives us the versatility to use whatever resources we have at any point in the game to make whatever we want. Currently me an my brother are in an area of the map surrounded by ludicrous amounts of literally every single resource in the game, we are going to stay in this area until manufacturing blocks are in COOP campaign. Here we are both constantly reminded of what we could be doing. We could be crafting, scrapping, automining, and refining our way to victory. But instead of plowing forward we're progressing at a snails pace picking up whatever scraps we can as we go.

This needs to change, I know eventually, it will change. How fast that happens depends on how important crafting blocks are to the development team, which at this moment it seems like it is an important issue to deal with at a later date.

I really don't like saying anything bad about the game, but I have to say it because I care about it:

CO-OP campaign feels hollow without manufacturing blocks, there's honestly no other way to put it. COOP campaign will never feel like campaign to me until they are available. I think this should be considered a high priority issue.

Now I really don't like showing a problem without giving a solution. If I were trying to prioritize several things at once and I didn't see a way to fix every block at once I would start small.

I would make a couple manufacturing blocks multiplayer-ready every update. I would add them in the order of how much it would benefit players in the current state of COOP.

For example, we currently have delivery cannons and receivers.

The two parts I would add in the next update are:
- GSO Pacemaker
- GSO Refinery

Now these two parts I chose for a specific reason. The pacemaker would allow us to speed up the selling of raw materials which we can currently acquire and sell in COOP campaign. The refinery would boost the income we make from the previously mentioned materials. A current bottleneck in COOP campaign is the speed at which we can sell chunks and the price at which we can sell them.

These two parts alone would make the wait for the other manufacturing blocks a lot more bearable.

If you finish those blocks, what next?

Choose another few crafting blocks and make them multi-player ready.

Maybe conveyors or scrappers.

Keep doing this over and over again. Over time you'll notice common issues with manufacturing blocks and you'll be able to fix them quicker, and you'll be able increase the number of blocks you fix every update.

Before you know it, you'll have fixed every manufacturing block in COOP campaign. Even the smallest of progress makes a huge difference over time.

I hope it's understood just how important the manufacturing blocks are to the game, and players like me, it just isn't the same without them.
 

Masakari

Well-Known Member
Joined
Nov 2, 2016
Messages
950
Likes
998
Points
505
Age
36
#6
Few points I was thinking about watching the vid since I wasn't able to be there during the stream.

1) One of a couple things not mentioned it seems is the fact that with planes soon (sometime later) to be a thing now, weapons will be even harder to use/more useless against them (unless you use missiles or the AA EXP gun maybe as that has not had a use to play around with yet). When they do make an appearance in game we will need a weapons overhaul to allow non missile weapons to target lead in order to fight back seeing as missiles are very rare to both appear in crates or trade stations. Yes we can craft them however in order to get the resources we would be under constant threat from airborne enemies and may end up losing any materials we had to go out to gather.

1a) Also so many weapons/toys to play with but we all end up using missiles due to the fact they are almost guaranteed to hit compared to 90%-95% of the other weapons (not including melee/short range weapons like shotguns or flamers). We could also use a couple more non missile AA type weapons. This also bring up the fact we need an AMS (Anti Missile System) or 5 (one for each corp maybe).


2) Another is the tech on screen icon pop-up and lock-on distance range will need to be increased to work with planes (and in turn to work better with other techs as well) if only to get our guns ready for the attack. This will also need to increase their activation/retaliation ranges (mainly for ground based techs and turrets to fight back)

Dunno about anyone else but I find it extremely difficult to use a plane against ground targets due to the fact that soon as I got a lock-on target to get ready to attack, I'm losing the lock-on due to how short the distance is while fighting the camera to keep my intended target in sight (radar can only get you so far in lining up with targets). I tend to have to make several to half a dozen or more fly-by attempts to line up on my targets and even then I miss my strafing runs most times (unless using missiles which tend to be quite boring after awhile).


2a) Speaking of fighting the camera, we could really use a/several FPV (First Person View) block(s) to better help us with flying. That or allow us to lock the camera or disable the camera spring/reset so it will stay where we angle it. Also a higher angle when on ground will be needed too to keep track or just to watch the skies for hostile planes.

The FPV blocks works amazingly well and stable in the time I have been using it (other than using R as the activation key which prevents me from using that key in control schemes). Its amazing how much they can change how the game-play feels like when using them (imo for the better).


3) AI Behavior currently is at best......viable(?) to a degree. Will AI get an overhaul as well before planes make an appearance? Few questions about how will AI deal with various things like currently.

--How will they deal like ZG above said with terrain when chasing us or set to follow with our own AI planes?
--Can they take off from ground?
--How will they follow us while we are in a plane?
--Will they try to bump and grind on us like now? (my personal favorite annoyance with them aside from auto righting ruining designs)
--Will they be the swarm of bees (drone packs) and cluster/huddle up when having more than 1 follow currently?
--Will they be able to fly in formation like you tried to show with various sizes/designs?
--Will they be able to use just throttled blocks (LME, HE/BF props/rotors)?
--How will they handle that 40-45 degree trigger of "Oh noes, I'm tipping over!" auto righting they do? Will that be a RMB control/radial menu wheel toggle (please...please make this happen) or because of it being a plane it will not be a function to the AI (you mentioned during testing that one liked to fly upside down)?


All I can think of right now.
 
Last edited:
Joined
Sep 20, 2019
Messages
22
Likes
13
Points
4
#7
I haven't seen the stream yet, but you used to be able to edit a file to increase render distance.
Dunno if it still works.
Really made a difference.

I think the render slider should go higher, like 4 times higher.
And have a tick box for 'yeah, i know this could mess up my FPS.'

Not everyone has a potato or console to play on.
 

ZeroGravitas

Breaker of Games
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#8
Thanks for your quick response!:D

Finally, the skybox and horizon also need a bit of work - the bright white in the day time is pretty glaring.
Yes, glare is a problem. Particularly if enabling distance blur - it seems like it adds gamma with the blur, or something (makes it kinda unusable).

I'd also wondered about the cloud patterns, having found out they are specific to the biome you are currently in - might need a more complex system once looking out over a vast patchwork of different biomes...

100-200 units. This is in the range of where we currently allow picking up blocks, but it would stop you from being able to interact with far away teleport towers.
Ah, right. Well, in honesty I don't use transport (quick travel) towers that much (but Ssundee has probably popularised them). But I have been moving back and forth between (mobile) fabrication base and harvester, etc, at almost max visibility range (well over 200m). And also important to be able to right click an allied AI 'plane' (i.e. hover) to set it to "guard" before it sails out of draw distance (with hover-bug forced movement). ((They currently revert to idle after player control, seems like, awkwardly.))

So, I'd personally opt for no world rounding unless you can apply some clever work around to (mid) distant interactions. I guess you guys might be about to allow tech switching and AI control via the tech manager menu (if not, please do ;)). That would mitigate a loss of distant interactions, but I'd still much prefer to have them.:)
 
Last edited:

ZeroGravitas

Breaker of Games
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#9
The squadrons of airplane AI seem to be crashing in unison, is that because they try to fly before the PED system is activated?
From what Sam was saying it was that the particular type of plane didn't work well with the universal PID configuration system he'd setup and all the formation followers faithfully went with the leader. (But I've no idea, it did look like it wasn't really even trying to manoeuvrer.)

if the AI gets advanced enough, wingmen following you around.
That does seem within fairly close reach, doesn't it, and it would be amazing! Not sure how players would get the AI wingmen up in the air though...? Do you just *have* to find a nice big open (salt) flat area to line them all up on? Or do they immediately pull themselves up into a tall build beam and start circling you upon spawning. (Both viable.)

Also, what about having AI that can transition from ground mode to plane mode? I have plenty of techs that are built to do this (immediately loved the idea of hybrid mode vehicles)... :)

I hope it's understood just how important the manufacturing blocks are to the game, and players like me, it just isn't the same without them.
Yeah. Good to hear others pushing on this point. I stopped, for the sake of more harmonious dev relations, heh.;)

The problem is that before co-op implementation of crafting would take a lot of time that means delaying other cool new features (maybe like these ones we're talking about, above.) And before that, there are also a lot of crafting system bugs to fix, possibly involving re-factoring (rewriting) much of the crafting AI code (which sounds like it isn't great). Which is why it's been left alone so long (a can of worms). But personally I think it's just something that needs to be done some time (like with having to implement an MP mode, because all games like this have one). I think TerraTech without crafting is kinda like a burger with no dressing or extras, just bread and patty. Not very inspiring and also it shouldn't be underestimated how much of a (good) time sink resource gathering can be for a large number of players - e.g. imagine Terraria without resource mining (or farming), just a side scrolling mob grinder.

But anyway, this has gone way off topic, so if you'd like to discuss (or push more for) crafting stuff, please post up a new feedback thread (and tag me there).;)
 
Likes: RC-3197

ZeroGravitas

Breaker of Games
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#10
we will need a weapons overhaul to allow non missile weapons to target lead in order to fight back seeing as
This is kind why I was saying to not have airplanes agro on ground-only techs. Similar to how Terraria Harpies only exist up at cloud level.

But yeah, movement compensation and target leading would be a very sensible tie-in addition, aslone with airborne AI (e.g. a new corp module).

his also bring up the fact we need an AMS (Anti Missile System)
Fair point. Missiles would be so universally used for air-to-air combat it would be way more one dimensional than the current (ground) meta... Without countermeasures of some kind. (Probably heavy blocks, so hard/impossible to use on small fighter planes.)

increase their activation/retaliation ranges
Yup, also essential point. It's been too so cheesy for so long that you can just missile enemies beyond their activation range that it might come as a bit of a shock to the existing player base. But this definitely needs upping (when AI is looked at, generally). Targeting lock too, for usability with planes. And have allied techs/planes actually target that target too?

That or allow us to lock the camera or disable the camera spring/reset so it will stay where we angle it.
Did you realise that sliding down to zero camera spring *actually* disables it, now? But yeah, the opposite might be helpful too, when camera is hard-locked to tech orientation (but without smoothing could be hard on eyes/brains).

General AI stuff - yeah, I'm interested too.:)
 
Likes: harpo99999

RC-3197

Well-Known Member
Joined
Jan 5, 2018
Messages
23
Likes
51
Points
415
#11
This also bring up the fact we need an AMS (Anti Missile System) or 5 (one for each corp maybe).
Fair point. Missiles would be so universally used for air-to-air combat it would be way more one dimensional than the current (ground) meta... Without countermeasures of some kind. (Probably heavy blocks, so hard/impossible to use on small fighter planes.)
Maybe they should implement my idea first suggestion, which was a missile point defense laser, that would be a really good missile defense. It was definitely thought of with airplanes in mind.

It's basically a passive 1x1 laser that intercepts incoming missiles. It can fire 3 times until it goes on cool-down for a couple seconds so if the enemy has venture missile launchers it's kinda useless unless you have like 50 of them on you. Also it would be really expensive, for balancing reasons.

Here's a video of what it is inspired by (Command and Conquer Generals Zero Hour):

And here's a video of planes with it getting absolutely destroyed so you know it's not OP:

But anyway, this has gone way off topic, so if you'd like to discuss (or push more for) crafting stuff, please post up a new feedback thread (and tag me there).
That's a good idea, I'll do that some time in the near future.
 
Last edited:

Sdarks

PAYLOAD STUDIOS
Joined
Jun 13, 2018
Messages
277
Likes
452
Points
270
Age
25
#12
There's a ton to cover here so I'll just work my way down the list responding to a few key points.


I'm really hoping that won't be too far off!
Looks can be deceiving here. While it may look like the flying AI is ready to ship next week as a full feature it's really not. It is nowhere near in a state I'd be happy putting it into an unstable. It needs a full design cycle with plans and proper implementation and it'll break a lot of things, so it may not be coming any time soon if at all. I hope we made that clear during the steam. Some of the reasons why I'll outline later on in this post.

Pure player-type control inputs, no physics cheats!
Correct! I even have a debug mode which I forgot to show off that directly takes control of the player tech.

) In the first half of the stream, we saw all sorts of ugly visual artefacts with the lighting inadequacies, etc. But they didn't seem noticeable at all in the second half (with the AI planes). Is that because you were deliberately provoking them to start with? (Over-extending render distance? Moving time faster, etc?)
Oops! So I restarted the game at one point to go into creative mode and that would have removed a lot of the tweaking that Django did to the camera and view distance. So it was probably using the old draw distance, or a mix of the settings of old and new.

That plane control code looked an aweful lot more involved than just PID, as far as I know that technique (from it forming the core modules of my Cybernetics degree)
That's because it's TWO PIDs! I'll elaborate later.

Presumably you have some kind of evaluation function to assess how well the AI plane is reaching it's desired trajectory. Or is that included in a code module or something?

Also, how are you even figuring out what trajectory to aim for? I mean, that's the big difficulty here, no? And how to roll and then pitch to be able to achieve these things seemed (to me, before hand) like it would be a very non-trivial AI issue....
I'm sure I can share a little bit on how it works. Because it's a prototype there's a lot of smoke and mirrors going on. The prototype works by having two PID controllers, one for pitch one for roll. The target pitch and roll are basically just the further away it is (angle to the player and height compared to player) the bigger the target (up to a maximum), a lerp with a maximum essentially (smoke and mirrors!). The magic of PID controllers behaving somewhat like machine learning (without the complexity) is that by combining the two they both kind of respond to the effect of the other and adjust their values accordingly. The smoke and mirrors comes in here with the maximum roll value, I found that if it rolled too far, because I had no rudder control to work with, they would just fall out of the sky as there's no way to maintain pitch. All of these are obviously things that wouldn't make it into a live version, as while game programming is often hacks and duct tape prototypes are another level. The idea here was to prove it's possible and discover the issues that will crop up should we try and implement it correctly.

How well do they cope, currently? Pressumably they are totally blind to terrain height? So would tend to fly into tall mountains? How tricky might it be to have them pick themselves back up off the ground? e.g. in a height extended build beam?
That's one of the things again that makes this a prototype. They can take off quite well on flat ground but have no concept of trees or terrain at the moment, so I just set them to fly above most normal terrain features. There's plenty of test videos I took of them flying into monuments, trees, big cliffs etc. There's some ideas for dealing with that (raycasting ahead of the plane to detect terrain and similar) it just wasn't worth implementing in the prototype. As for the build beam bit that would definitely be a design decision. Perhaps we could have them build beam and rotate to a direction with less trees? Or perhaps the mega build beam floating above the trees is the route to go?
 

Sdarks

PAYLOAD STUDIOS
Joined
Jun 13, 2018
Messages
277
Likes
452
Points
270
Age
25
#13
The planes that you showed off were all extremely light weight, high wing ratio and very manoeuvrable. Did you test out heavier, chunkier (and far bigger) planes, too?
I absolutely did! That's part of the reason for doing a prototype, to see what kind of planes we could have. I think it's just coincidence that the planes that showed up were the smaller ones. There's a couple of quite large cumbersome planes I also tried out that actually worked better. I actually had to add some passive gyros to some of the aircraft to reduce the manoeuvrability as the AI were over compensating and doing too extreme movements to maintain stable flight (more smoke and mirrors!). Here's one of the snapshots I had in there I was hoping would show up, courtesy of @Flan


Flying Bertha.png

Yours all had propellers, what about boosters (and hybrid fan/boosters)? Think it would be hard to code them to use that in a semi-intelligent, sparing manner?
No boosters! If this ever gets implemented properly? Don't think it would be an issue. For a prototype? Too much effort.

What about helicopters and anti-grav aircraft? I'd imagine that wing-less AG would be relatively straight forwards (with even simpler PID control). But throttled lift craft, yikes...
That's definitely a different kettle of fish and I can't even begin to speculate. Let's figure out planes first :D

It seems like enemy spawning would certainly have to be overhauled to coincide with (both of) these features. While it would be cool to see *some* planes appear while player is entirely on the ground
Absolutely, again one of the reasons why we stress this is a prototype and will not be shipped anytime soon or in this form. There's a lot of work that you can ignore and skirt around with a prototype to get something working quickly and population/spawning was absolutely one of those. It just spawns planes randomly around the player every 30 seconds or so.

But also the de-spawn range would have to be significantly increased - looks like you had to do that for this demo, and waived the 200m immobilisation limit, too
Yep, I had to put some hacks in, it would need to be investigated for a proper implementation.

I'm glad that a guy who worked on flight sims is working on that as nobody else in the world would do a better job than him
Thanks for the compliment, but trust me there's a lot of people in the world that would do a better job than me on this one :)

The squadrons of airplane AI seem to be crashing in unison, is that because they try to fly before the PED system is activated? That's what it looks like to me.
I'm not actually sure why they were crashing so often with that one type of plane. I didn't see that while creating the prototype. It's one of the joys of live demos!

AI gets advanced enough, wingmen following you around
There's some added difficulties there as the player is a lot less predictable than an AI to follow around but it's potentially possible.

AI Behavior currently is at best......viable(?) to a degree. Will AI get an overhaul as well before planes make an appearance? Few questions about how will AI deal with various things like currently.
This along with a lot of your other questions would definitely be things to ask if and when we actually decide to push forward with flying AI properly. We don't know if we'll rework AI at all, or before this, or after this, or anything like that because this is just a demo to show what flying AI might look like. Same goes for all the "will they" questions as we don't know! We haven't designed flying AI yet or agreed to implement it at all so we have no idea what it might look like. Part of the reason for showing this off is to get an idea from you guys what you'd like it to look like if we did implement it.


I've intentionally skipped over a lot of the questions about co-op crafting because that's absolutely not my area of expertise and gone a bit off topic, so don't think as a company payload is ignoring you just that I personally am not the one to answer those questions so haven't answered them. Same goes for the weapon balancing, that's very much a design discussion and will definitely be a thing to consider if we add flying AI properly.

Glad you guys enjoyed the stream and if you have any more questions (or I missed any important ones) feel free to tag me.
 

ZeroGravitas

Breaker of Games
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#14
It needs a full design cycle with plans and proper implementation
Yeah, that doesn't sound too bad. Attainable (though never immediate). Certainly it could form the core content of a 'major update' and make quite a bit of buzz.:)

I even have a debug mode which I forgot to show off that directly takes control of the player tech.
Oh cool! Give us a screenshot of the UI or little gif sometime.;)

So it was probably using the old draw distance, or a mix of the settings of old and new.
Well, you stayed in campaign for a while, then went over to creative. But in both the draw distance stayed *way* beyond the normal max (~600m during daytime). e.g. in creative (around 1:04:06) I could see 3 monuments at once and the terrain looked pretty pristine, whatever the settings, do that (maybe with glare filter) ;):

Cropper2019-10-11-14-50-19-4804943.jpg

Current (night time) view distance comparison (lol!:p;)):
Cropper2019-10-11-16-25-05-4756683.jpg

Your part of the stream did show tile illumination discontinuities, most visible at night on the salt flats (before that), but that's just like what we already see in game, e.g. above (and reported previously):
Cropper2019-10-11-15-10-11-0393553.jpg

The prototype works by having two PID controllers, one for pitch one for roll.
Oh, that's far simpler and more elegant than I'd imagined. Smart.:oops::cool:

So the target is the player's tech (or a spot directly above it, if it's too low). And the the angle, in the horizontal plane, between the AI plane's current heading and the target is determined. The further off course it is, the bigger the roll angle it aims for (i.e. leftwards, if the target is on the left, etc). Up to a maximum (definitely never more than 90 degrees). The PID control just determines how 'hard' to press roll-left to try and reach that amount of roll fairly quickly, without overshooting or oscillating. (Do you have 'pulse width modulation' on regular binary key input?)

Once the plane is rolled over, the apparent pitch angle between the plane's heading and the target will increase, stimulating the second PID to pitch up (more strongly), pulling it 'up' towards the target (from its point of view). Kind of an emergent behaviour. Nice.

Ce?

We don't know if we'll rework AI at all, or before this, or after this
Not to be finicky (would I ever...?) butthe roadmap blog from 14 months ago stated as a "first priority post launch":
Improving core AI behaviours (Follow, Harvest etc)
I always assumed that was hyperbole, but c'mon you guys can't just not fix the AI up at all, a bit...:confused:

For example, the kind of way you'd allocate 'wing-man' slots for planes in the air would also be applicable to ground followers. But probably more of a duckling configuration behind a ground tech, with a string of virtual slots each big enough for the tech's size. (Rather than as current, where all Ai piles into the same spot at/inside your rear end...)

the player is a lot less predictable than an AI to follow around but it's potentially possible
In a full solution you'd possibly want more PIDs, e.g. to try to match airspeed, too, for formation flying. (Easy enough to have a Yaw PID too, in parallel with the roll control, right?)

How did you code your formations? Their spacing kinda looked too rigid to be just targeting a spot besides a lead plane...? (Still no physics cheats there?)

With your upside-down Ai plane anecdote, is there any chance it was cause by [A] and [D] keys being switched in the plane's control scheme? Quite common cocking those pairs up. (Would that produce the behaviour? Off-hand guess.) And is that before/without a maximum roll angle limit?

Anyway, a common problem with TerraTech plane designs is a lack of tail wings, such that the tail of the plane looses altitude if you roll much, stalling it sideways and falling out the sky. (A Yaw PID could help counter that, provided there's *some* tail plane.) Also just loosing a lot of altitude on turns.

Here's one of the snapshots I had in there I was hoping would show up, courtesy of @Flan

It's fairly heavy, but also very well well powered, fast (~140mph), short, balanced and with wings that give sideways lift.

What about GSO planes, like these? (With ~80-90mph airspeed and heavier per wing area.):
ZG Plane Tube 1h.png
ZG Cluster Dropship.png

Oops, next post (too many attachments)...
 
Last edited:
Likes: Aardvark123

ZeroGravitas

Breaker of Games
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#15
Or very dense, but faster, like this:
My First Tech.png

Or very large (huge would be very fun to see!!:D):
The Fallen God.png

Or a flying base (with vertical fans assisting lift):
GK 工业飞机.png

Or conversely extremely 'lifty' and ultra manoeuvrable:
Streamlined.png

I don't really have any bad plane examples to hand (but then you guys just wouldn't put any in anyway, so no worries, perhaps).

And is your AI flexible enough to just rebind the propellers control in the tech's scheme to also hit boosters? Or forwards acceleration? e.g.:
HvrStr Jet 2d.png
(Also switch roll for turning with this one, I think...?)
 
Last edited:

Sdarks

PAYLOAD STUDIOS
Joined
Jun 13, 2018
Messages
277
Likes
452
Points
270
Age
25
#16
whatever the settings, do that (maybe with glare filter)
Not sure what that is, guess it's some of Djangos changes with the atmospheric scattering still up to full or something. Everything looks better at dawn and dusk though :)

So the target is the player's tech (or a spot directly above it, if it's too low). And the the angle, in the horizontal plane, between the AI plane's current heading and the target is determined. The further off course it is, the bigger the roll angle it aims for (i.e. leftwards, if the target is on the left, etc). Up to a maximum (definitely never more than 90 degrees). The PID control just determines how 'hard' to press roll-left to try and reach that amount of roll fairly quickly, without overshooting or oscillating.
Pretty much spot on. They do oscillate a bit though, it's inevitable unless the parameters are perfect.

(Do you have 'pulse width modulation' on regular binary key input?)
We have proper axis control from consoles so it just uses that, imagine it's using the left analogue stick only.

Linear Interpolation fancy word for something pretty simple.

Not to be finicky (would I ever...?) butthe roadmap blog from 14 months ago stated as a "first priority post launch":
Things can and do change obviously. Not my place to comment on the whats and the whys so I leave everything as a vague "who knows" and let @Matt comment on anything more concrete when it's appropriate.

Their spacing kinda looked too rigid to be just targeting a spot besides a lead plane...? (Still no physics cheats there?)
No physics cheats, I did cheat a little though. They match their squad leaders inputs mostly, until they get out of line and then they try and fly back to their target spot. This is why they turn and you just get the whole formation flying sideways instead of the V shape turning correctly.

As for the snapshots I'll try a couple out, but there are a few issues at the moment, things like it struggling with planes that have the propellers too far from the centre of gravity (it'll start using them to turn) and things like that.
 
Last edited:

Sdarks

PAYLOAD STUDIOS
Joined
Jun 13, 2018
Messages
277
Likes
452
Points
270
Age
25
#17
With your upside-down Ai plane anecdote, is there any chance it was cause by [A] and [D] keys being switched in the plane's control scheme? Quite common cocking those pairs up. (Would that produce the behaviour? Off-hand guess.) And is that before/without a maximum roll angle limit?
Nope the controls were fine (it doesn't actually use the control schemes it just works only with planes that work with the aeroplane control scheme). I believe at that point I was treating upside down as 0 roll as well as right way up as 0 roll, so it was purely that the plane flew better like that. I think it just rolled upside down naturally and it was easier to keep it level like that for the PIDs
 

ZeroGravitas

Breaker of Games
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#18
I'll try a couple out, but there are a few issues at the moment, things like it struggling with planes that have the propellers too far from the centre of gravity (it'll start using them to turn)
Cool.:) That was an issue for (player controlled) planes at one point after the new schemes, but I think propeller control now overrides pitching and yaw control of vertical fans. (But off-centre horizontal fans still stall out for helicopters, awkwardly.)

... But then does your AI interface at a lower level of control implementation, so still has this issue with planes?
 
Likes: Aardvark123

Sdarks

PAYLOAD STUDIOS
Joined
Jun 13, 2018
Messages
277
Likes
452
Points
270
Age
25
#19
Here's some examples of the flying AI controlling the player tech with various snapshots. All I'm doing here is setting the target pitch and roll off screen, not using the keyboard at all.

Big plane

Another big plane

Weird small plane

Plane it really doesn't like

Crazy little steering hover tech

As you can see it works with most of them, but there's some it doesn't really like. I think the main issue is with the ones that have vertical facing fans as it tries to use those to pitch which doesn't work so well.

That's all you're going to get in terms of snapshot tests though, as I have other things to be doing. I think it gives you a rough idea of how it's working though :)

Edit: The low FPS is on the GIF not the game, sorry!
 

ZeroGravitas

Breaker of Games
Joined
Jun 29, 2017
Messages
3,798
Likes
5,227
Points
675
Age
36
Location
UK
Website
www.youtube.com
#20
Oooh, thanks for the additional tests. Interesting! Love all the little planes zipping past the 'Fallen God'. It starts to look like a flight-sim game, with the landscape spread out before you like that too... :D:cool:

As you can see it works with most of them, but there's some it doesn't really like. I think the main issue is with the ones that have vertical facing fans as it tries to use those to pitch which doesn't work so well.
Ah, right, well if you can see those fans revving up and down then I take your word for it.:) Otherwise, from what I can see, it also looks like the PID has too little D (differential/dampening) to account for the high rotational inertia in the roll axis, leading to increasing oscillations. (With the factory plane.)

Looks like also a finished plane AI would have to limit pitch in the vertical axis, specifically (against gravity), to conserve airspeed. As that factory plane pretty much stalls out the sky too (before the roll oscillations). I felt that was likely to happen with the GSO fan propelled techs at the end of my post above that, given their lower air speed, easier to stall. Also prone to over rotation with roll (tube plane particularly). But interesting that the "Weird small plane" (fighter jet with GSO armour plates) flies OK - it's dense but has quite high airspeed.

Oh, btw, what location does the AI target when controlling player's tech (e.g. in those gifs). Nearest enemy, when one's in range, or otherwise...?