#gameplay-ai
1 messages ยท Page 175 of 1
Oh damn. I hate everything about about pitching. It always feels like "good" pitches are about making vague promises. Glad I only see that side from people who need some data/ideas from me.
is the process a pain in the ass? how much writing is involved in these grants
Yeah, pitching is the pits.. I've got one on Tuesday
"pitches are about making vague promises" yes
lol
sad but true
Depends on the funds.. smaller grants of like < 30k are relatively painless for the most part
This one on tuesday is for 12k, basically a 5 minute slide deck and a few meetings
And I got to use DallE to generate the art for the slide deck.. for the funzies
Yeah, the ones I've been involved with have mostly been in 300k range lately
hahaha nice! I should try that
last one was 2.8 mill though, but we are only going to get 1 in 100 of those
Or even 1 in 200.. don't know the accept rate right now
I mean I'm kind of one of the few doing defence meetings.. so I get put on all sorts of hopeless bids ๐
Part of networking as an academic, is realizing that most bids don't get funded, so you have to do a load of them
yup
this is the trick
just keep trying until someone bites
lol
the shitty thing is
Well organized research groups have paid employees doing that bid prep fulltime
sometimes some of opportunities are reserved for certain companies/individuals
so even if your idea is amazing you migth still not win
which sucks
we have a few haha
Yeah, I was in a defense briefing last week with exactly that feeling..
some people that purely focus on writing these bids
Yeah, sadly I don't have that.. so some of our bids suffer
Mostly due to lack of time or info or practice for the specific funding body
They all have quirks
yup! each agency is very nitpicky about what they want
some are super tech saavy and want all the math formulas
others just care about the end result
What stuff they won't fund, or expect match funding for
haha
What they regard as match funding.. oof ๐
My fave is the "light touch, just get it done" variety, but thats rare
this is crazy to me too right. Like getting funding is already hard enough and then they want someone else to promise to double the fund hahaha
finding one source of funding is hard enough, where you gonna get two
Well, most of the time, you can claim staff time as match funding, for some funds at least
Which basically means my/our time gets booked on it
i wonder if that applies for academia only
for us they expect an actual entity to sign an agreement to provide the funds
We get those too.. it depends on who is the funder.. the one in question was an EU fund
which we don't get access to anymore ๐ฆ
damn :/
have you guys tried applying for Unreal Grants before?
have any experience with those
Yeah
did you get funded?
I've gone for one, but honestly it wasn't strong enough
But the other one was a solid one.. about creating a virtual production mobile to take to schools
Weirdly, the first one I put in, someone else got funding for ๐
interesting, wonder if the timing wasn't right or something
The VP mobile got a bit more play than mine, but Epic weren't exactly on fire responsive and the pandemic kind of knocked a lot on the head physical contact wise ๐
ahh what a shame, the idea sounds intriguing
Well, all I can say, is that the ones who got the funding, still haven't delivered a project.. I bloody well would have ๐
hahaha
i thought about applying for one, for continuing some work-related reserach but never got around to it
I've got a small loss leader commision to do some simulation visualization stuff for a navy training thing
But I was already playing with some of it anyway
Tiny funds, but enough to buy some art at least ๐
how you'd go about getting that commision, they just kinda reached out?
A long chain of networking
haha
This is in fact, to try and get more paying stuff
Honestly, its borderline insulting, but these guys have an in with more people
So we keep everyone sweet and hopefully it pays off down the line
Yeah, its all about that really.. knowing when to not do things and when to speculate a bit
The Uni wants us to chage like 1200 gbp a day ๐
1200-1400 ish
But not realistic for a lot of things
So as an academic, I'm allowed to do 10 days consultancy and pocket it for myself
10 days per month, year?
OR, I can do consultancy and split it with the Uni
Or I can give it entirely to them
10 days a year
Yeah, exactly ๐ aint nobody interested in that!
haha how much actual availability do you have given that you have to teach so many classes?
For funded grants though, its different
Well, so with grants, we can buy out our time
ahh
i.e. give the uni money, so they can hire someone to fill the gap (theoretically)
One bid we did recently, I was down for 1-2 days a week on it, for 3 year project
Ahh interesting, so there's some flexibility there. That's a nice little perk
Well, you have to negotiate it with your boss
so this is where having a good manager comes in
That said, I doubt any of them would turn down money, political suicide ๐
it's good to learn how it works on the other side of contracting with academic partners
so
What they really want, is for us to win like 3 million funding, where PhD students do all of the work, where academic staff get a tiny allocated time for supervisions
do you actually get to work on the projects, or do you assign some students to work on them?
yup
this is exactly what i was thinking lol
this is sorta how it works for us too
Usually students, but I'm more of a hands-on guy ๐
Plus coming from games, I'm the low guy on the academic totem pole
yeah i would want to work on the projects myself
so I deal with tiny grants and loads of networking and much of the work myself
aaand a bunch of teaching to boot
Not ideal
But the main advantage to academia, is autonomy
My boss has basically 0 idea what I do ๐
We had a research presentation week last week.. everyone put up posters of their research.. it was quite telling that mine was all AI generated art and everyone else was like insect killing robots and such ๐
or self driving farm robots
or.. healthcare robots
Basically, nothing even close to being creative.. apart from a colleague showing off her game ๐
haha, you stood out
interesting
this was all hosted within the university itself?
I realized that it wasn't my crowd honestly
Yeah, it was the college of science research showcase
COS has like maths and physics, geography, CS and games, engineering etc
Luckily, the arts show is soon and psychology is also soon.. both of those will be fun
We in games tend to do more with those rather than.. "science"
Naw, this was in person in our engine shed
ahh got ya
Basically an old train shed that got turned into a concert venue
Its meant to foster collaborative research
But of course doesn't come anywhere near ๐ for me at least
so its common for some professors to partner up with other professors if their individual research aligns with each other?
haha but i assume the art department would love your research
Well, usually its more inter-university
ahh
I'm hoping to do a gallery over summer
I suspect it'll be fairly controversial ๐
Most large grand funding goes to consortiums of University + businesses
grant funding even
so like 3-5 million
The point is, that the research is funded, then commercialized via the companies
But you might have like a dozen Universities on the project
that's very similar to how we go about doing things as well
So each gets a cut of that grant funding.. usually for a PhD or two
given your research, do game companies every partner up with academia to use some of their ideas?
i haven't heard if that's the case
only very rarely
I know at least for some of our larger projects, we have used game studios before for building some simple systems
games companies are honestly, just not that innovative
I've spoken to a few that dismiss ML outright
So myopic
It partly comes from hubris about their own superiority
But also, most games execs are actually quite clueless and often pretty conservative
Its weird

I've found that people in the fashion sector are waaay more innovative ๐
I was already iffy enough about gaming industry, now im even less convinced to join
ai/ml in fashion industry?
I think part of it, is just that games doesn't make enough profit to take risks
ohhh what are some of the use cases on that industry?
yeah, loads of AI/ML in the creative sector in general
So for instance, virtual try-on of clothing
Or generating designs for new cloths
Or doing physics simulation using ML
yeah very true, i recall a friend interviewing for EA. And he was clearly told that on (i think FIFA franchise?) he wouldn't get to innovate or create any features, it was just about maintaining what they had for the next sequel lol
Or just predicting markets
huh, that's a pretty interesting application. I should read up on that
I mean, they're all set for the metaverse when it comes along
So imagine the virtual branding stuff you could do with virtual clothing designs
and translating those into real-world clothing sales
Sadly, this will all be tainted with NFT's and the like ๐ฆ
yeah i hadn't even considered those application,
I've got a postgrad student doing a project over summer looking at photogrammetry vs neural radiance fields, because I want to build a clothing scanner to support it ๐
what are neural radiance fields?
they're a form of ML based graphics
they define the boundary probabilities of a mesh surface
kind of like signed distance fields if you know them?
yup
one sec.. I'll link a vid
im looking at some images for neural radiance fields
When the first instant photo was taken 75 years ago with a Polaroid camera, it was groundbreaking to rapidly capture the 3D world in a realistic 2D image. Today, AI researchers are working on the opposite: turning a collection of still images into a digital 3D scene in a matter of seconds with the implementation of neural radiance fields (NeRFs)...

It functions similarly to photogrammetry
that's awesome
but the representation is fundamentally different
But the point is, you can generate meshes using them ๐
I wonder how much cleanup is necessary to get these game-ready
Part of our study is to measure failure cases for photogrammetry vs NeRF
very cool, do you guys publish your findings anywhere
Depends if the student gets their act together ๐
lmao
Hoping to get a postgrad to do an Unreal Engine RL project this summer
one of my better undergrads (the one who did the Unity ML agents project)
what kind of policies you're trying to learn?
i haven't read the paper on GAIL
This was mostly just evaluating the toolset for his undergrad.. seeing if he could figure out how to incorporate demonstrations into the RL approach
ahh I see
If he decides to do it, we'll likely build an RL agents API similar to the guys at Embark Studios
Run it in the cluster
Kubernetes/Kserve etc
What kind of work did Embark Studios do?
Lemmie link you the video ๐
ahh is it the agent learning to walk on complex terrain?
Check out the world premiere reveal trailer for Arc Raiders a new game brought to you by Embark Studios unveiled during the 2021 Game Awards livestream.
Yarp
the robot spidery thingy
near the end of that trailer
Embark are a really interesting studio.. doing mostly PCG AI/ML based content production
ohhh yeah i remember this, i had seen the prototype blog post about it but didn't realize it was actually in a game trailer lol
Worth looking at their presentations
oh yeah im gonna look it up
innovation + funding, where do i sign up
They did some presentions with Nexon, who I think own them now
mayube they are hiring
You know speaking of this GAIL approach, i've been looking at research for buildign more sample efficient models
in my feed i got recommended this
Project page for Jump-Start Reinforcement Learning
looked interesting
Nice
I'll pass it on ๐
Although he might have cited it already
it looks familiar ๐
haha wouldn't surprise me
im gonna go have lunch, good talk!
see ya
In this complete reinforcement learning course you will learn everything from implementing double Q learning and SARSA with just numpy all the way up to implementing Deep Q Learning and Policy Gradient methods in Tensorflow.
No prior knowledge is needed, other than basic proficiency with Python.
This was cross posted with the Free Code Camp c...
Knew I heard of it ๐
probably not the same thing ๐ hahaha
Take care.. ttyl
hey im stuck with something with AI movement, what i want is an ai to chase me fast and slow down when it gets to me and slowly approach me and then if i run away start to chase me fast again
right now the ai is chasing me fast and then it slows down and starts walking towards me but I cant get it to exit/abort that "moveto" task so it never speeds up again
I noticed that if I uncheck "Track Moving Goal" the ai will start to run again but unchecking that also makes the ai walk towards me in increments basically walking to wherever I was when it started the task
I'm trying to get the ai to exit its moveto if I get too far away from it so that I can adjust its speed to make it chase fast again
you can use a decorator to check how close it is to its target to abort the node
I have a decorator checking for a value thats being changed in an eventick in the ai's eventgraph "DistanceToEnemyActor", the value is being updated on a tick and i've verified its working
it looks like the decorator is only checking after the move task is complete and if i use "track moving goal" then it never completes
if I dont use track moving goal then the blackboard based condition works and is applied
so the value is set and the condtional is working as intended, its just that with track moving goal it doesnt check the condtional again
but without it the ai walks at me in segments to wherever I was at the start of each segment
The only difference between having that on or not is what it says - track moving goal or no
if you have it on, and you move somewhere else while it's moving towards you, it will automatically update its move target during the move to match
If you have it off, it will not update. It will move to the location you asked when the move started, and it will finish the task.
I'm hoping to not have to choose between the ai tracking movement and being able to abort the the move task
none of the options affects your ability to abort it
all tasks can always be aborted
Is there anything I need to get the AIPerception working besides the stimulus on one factor with soght enabled, and sight on a sensing component on another?
I did these and set the stimuli as auto register, but it won't pick it up.
I hate python SO MUCH ๐
Unrelated.
Yep
Well. I'm still looking, I did move the AIPerception component to the AI controller instead of the pawn, but that didn't do anything
haha let me guess, issues with libraries compatibility because of different versions?
what unreal version are you using?
5.02
I recall there being a setting where you set the types of actors you wanted to perceive. The options were like neutral, and some others. Is that still in 5.02?
Nah, just general hatred. I guess mostly just having to read people's shitty python code that is almost always half assed.
as part of the aiperception settings
hahaha, it does lend itself to that
I don't think so. I will double check
im checking right now too
I would think the stimulus component would be enough though
it's this stuff here
I recall there being something weird where you had to enable all of them in order to get detection due to some blueprint limitation. But i think that's been fixed
I don't recall where you specify where your actor is an enemy/neutral/friendly, maybe someone else knows
In 4.27 you still have to Detect Neutrals at least.
really?
i thought i read in the patchnotes that they had changed that, maybe it was for 5.0
Good catch. Thank you.
That would definitely make trying out Perception easier, if they aren't going to expose team attitudes to Blueprint.
If you want to enable team based perception, this thread has a good intro: https://forums.unrealengine.com/t/how-do-i-use-the-ai-perception-teams/120837/2
Well. Maybe in the future, I was trying to get it just print something if it detects the player, which it did, spamming it and thus lagging it I believe. But the main point is I have it working now
nice ๐
Yeah. I'm working on something for a GameJam entry.
can anyone help me out with ai in behaviour trees
basically i need it so when its chasing the player (which is what ive done) it plays from a walking to a running animation i have and it speeds up. The idle/walking and random locations work fine. just that
This sound like you just need to configure the movement component of your AI
Try checking out the acceleration of the movement component
this is what i mave rn
have*
when he chases the player i basically want it to play the running animation instead and speed up
but when hes not chasing, to go back to roaming
Agree with Josue, you probably want to use two things: a) the Movement component of your pawn/character to control accelleration/max speed, and b) your AnimBP to control blending from your walk anim to your run anim based on velocity.
i see yes. the first way ive done that but it has multiple animations so it will stay at that max speed of 150 (which is what i set it too)
and also with the 2nd way, when i add the running to the animation blueprint, what do i do then?
this is my animation blueprint
I'd suggest maybe simply making a new BT_Task where you re-assign the movement speed of your character
this is the other part of it
btw, i am new to game making and i want to learn so sorry for being so dumb ๐
That looks pretty reasonable (I'm not super good at Animation BP's, honestly).
i was thinking of that, but im unsure how to do it
But you're using the tools I would: getting speed from the movement controller, and using a blendspace to determine the animation.
ik how to make a new task but like im unsure how to do the speed and animation wise
seen as im new these words are so confusing ๐คฃ
you won't have to touch the animation
I was where you are 18 months ago ๐
if the animation blueprint is implemented correctly, it'll adjust by itself
you just have to update the movement speed and the rest will take care of itself
im very lost ahaha
it should be
So does you character never show the running animation?
could you explain how i would make a new task updating the movement speed and adding it to my behaviour tree ๐
not at the moment
If that's the case, you will just need to make sure your max speed is fast enough to show the run state.
if i add "play animation" then select the animation BP it goes.. weird
it plays half of the walking and half of the running and moves slowley
ahaha
you can modify 'Speed' in your animation blueprint to test out and make sure the animation actually runs
the one in this image
where can i turn up the speed again?
Bottom right, where it says speed.
try manually typing 600 on there?
goes back to 0
and then attach it to the chasing sequence?
oki ๐
when you make that variable NewSpeed
ill try that too
make sure you make it InstanceEditable
oki
but idk why the speed wont change
then in your behavior tree just make sure you set the value to something else, like 150
just receieve
oki
I wonder how much better AIPerception is performance wise to what I jury-rigged in Unity for AI,
In that I made a box trigger, rotated it 45 degrees, centered it to their eyes, and did a linecast on tick for whenever something was in the box that should be registered as seen
Sight Perception isn't doing all that much different from that. It's just smart about deciding when to retest vision.
i've done that too, just with a rectangle instead lmao
@hallow flint
Well. I wonder how fast AIPerception is compared to that.
it's definitely not updating the speed
Deciding when to retest I guess is the difference
yeah
yeah i think for my jam i might just attach something to my AI's face to act as perception, i dont need anything fancy lol
you forgot to check off the 'Success' mark
OHHHHHHHHHHHHHHHHHH
shoot i did
hold up
haha
you see the little running rhombus
that you have near the center?
the one for running
should be allll the way to the right
im not sure where they are placed
but
idle should be all the way to the left
walking should be in the center
running all the way to the right
ok
so
we fixed that
sorta
the walking is now mixed with idle LOL and when he spots me he dont speed up and dont run
im sending video now
he seems to be walking slightly faster
just slightly
i think i may know the issue. the snippet of code i sent you for speeding up may be wrong
try increasing the speed to like 900 and make sure it works?
okay let me check something
oki
also if i put the walking like dot thing on the animation BS then the walking works just fine
like its not "mixed" with idle then
yeah putting it in the right place smooths out the animations
wait
dont quote me on this
but i feel like the issue is
i have a feeling he isnt actually chasing me

but
๐คฃ
the only issue now
well i say "only"
but he doesnt return to the normal speed and back to roaming/walking after seeing me
are you calling the setmovespeed btttask on both branches of your tree?
yeah that one
i think
it works i think
when i close the doors tho he just starts running to the other side where the door is open ๐คฃ
yeah he's just pathfinding around the obstacle
any way to just like, prevent that? like when he gets blocked he goes back to roaming
so when the door is closed he just returns back to roaming
uhhh probably with some aiperception
like the stuff that @final loom was just asking about
or are you already using that?
um
i dont think so?
im sure ill find a fix soon
while im here rn, is anyone any good with jumpscares when the ai touches you it plays the jumpscare? with behaviour trees
What way? Like an image over the screen, or an animation?
and thank you so much @hallow flint for your help you are and all of you are amazing
animation
which i probs can create using ue4 but how would i do it with behaviour trees
just basically when the ai is chasing you, if he touches you it plays the jumpscare
Probably set a bool in an animation blueprint and have it set back to false after the animation is done playing
Well.
You'd need a reference to your animBP in your AI's class, or controller, somewhere it would need to store the reference
how would i like reference it
Well. What I do is create a var to the type of anim BP, then get anim instance on begin play, cast it to said type of anim BP, and voila! There's your reference.
Yeah, for your class of animation BP
Thanks again Josue for helping me with that issue. Would have been a while if I noticed what you said, if at all.
I recall getting stuck similarly to that before, but forgot what the fix was.
Ahh you could create a sub-tree off of your chasing branch. That basically checks for distance between your ai and the player character. If it reaches a distance of let's say 150-300 then you just run a playAnimation task. I think that should be it
no problem, glad you got it working
The most important part anyways
My EQS test which filters out items doesn't do the test for every item. I assume it has something to do with performance? That the test is to heavy and gets discarded after a certain point? Is there a resource to read about it or can someone guide me on what to look at for that?
That isn't a thing
If the item is filtered out by another test earlier on, then the rest of them might not run for it I guess, but perf cost doesn't affect it because that could produce extremely unpredictable results
it's timesliced, so a slow test just means it will not be able to run further tests on that particular frame
i dont get it why the delegates name is "onqueryfinished" if it has a "processing" member as enum 
It's possible that the same delegate type is used for multiple different events
Maybe it makes more sense on others
member variable name is OnQueryFinished in UEQSInstance
and only get called once, with one of the enum members
does not get updated
bad naming 
Yeah but that's not the delegate type
wouldnt it make more sense to name it as OnQueryStatusUpdated or something?
Not if it's called only when the query finishes
But if it's finished, why it is Processing?
Hmm yeah looks like it has its own delegate type so my theory is wrong ๐ค
FORCEINLINE bool IsFinished() const { return Status != EEnvQueryStatus::Processing; }
FORCEINLINE bool IsAborted() const { return Status == EEnvQueryStatus::Aborted; }
FORCEINLINE bool IsSuccsessful() const { return Status == EEnvQueryStatus::Success; }
FORCEINLINE void MarkAsMissingParam() { Status = EEnvQueryStatus::MissingParam; }
FORCEINLINE void MarkAsAborted() { Status = EEnvQueryStatus::Aborted; }
FORCEINLINE void MarkAsFailed() { Status = EEnvQueryStatus::Failed; }
FORCEINLINE void MarkAsFinishedWithoutIssues() { Status = EEnvQueryStatus::Success; }
FORCEINLINE void MarkAsOwnerLost() { Status = EEnvQueryStatus::OwnerLost; }
This explains it a bit better I think
Not sure why the finished status is called "processing", but it would appear so
But yeah looks like if the query is aborted or fails the status is different
You shouldn't crosspost your questions unless you're not getting answers, because I think this was now answered in #cpp
sure sry
Can I ask. How could I do that๐ญ
So you should probably make a decorator that simply checks the distance between your AI and your player
so add another decorator on my chasing branch on the tree?
with that being a blackboard decorator?
how would i make it so it checks the distance between the player?
Yeah so add it to your chasing thing
The logic for distance is just something like
(AiWorldLocation - playerLocation).Lenght
Subtract the the location of both of them, then get the magnitude (length) and then check if your AI is basically on top of your character
So a value like distance < 300 might tell you that he's basically on top of you
Or alternatively you could check if your AI is colliding with your player
yes thats what is probs better
cuz he would actually be in your face then before jumpscaring you
how would i make it so it checks if the ai is colliding with the player
sorry if your busy
but ive added the decorator
idk if its the right one
@ocean wren maybe interesting for you https://twitter.com/halllo/status/1536682463422038016
Basically "lets not become China" act?
guys
how do i make it so the behaviour tree doesnt start for a certain amount of time
so i can have them on the stage area for a certain amount of time
ive tried using the "wait" but it will just repeat the wait which isnt what i want
like is it possible to make it a "one time" wati
wait*
@hallow flint
sorry for tagging so much
So I wouldn't do it like this. I'd remove it from that sequence node and add an additional sequence node.
There you'd add this new decorates
To switch between chasing player or scaring player
Dark pattern AI, I don't think I've ever heard of that
Why don't you add a delay on begin play before you actually call the behavior tree
where would that be again
i dont think its ever "called"?
like this? @hallow flint
Can smart object claim handle be reused? I'm implementing a behavior when the player can start a dialogue with an NPC and I want the NPC to keep the claim on a smart object slot it's currently interacting with. So I stop the gameplay behavior without calling USmartObjectSubsystem::Release, but after the dialogue is over and the BT execution gets back to the task with USmartObjectSubsystem::Use with the claim that I stored for NPC, it fails and logs
Ensure condition failed: SlotState.GetState() == ESmartObjectSlotState::Claimed [File:D:\build\++UE5\Sync\Engine\Plugins\Runtime\SmartObjects\Source\SmartObjectsModule\Private\SmartObjectSubsystem.cpp] [Line: 471]
Error LogOutputDevice Should have been claimed first: Object:845242607 Slot:6 User:1
It seems that the slot changes its state to occupied so is the only way to keep the slot for the NPC is to release the slot and immediately claim it again?
hey folks is pawn effect on nav mesh?
the with deer wanna go to that bush
but can't cause body block from another deer
how can i fix this?
try using crowd or rvo avoidance
what is this?
options in character movement component
google ue4 crowd avoidance or ue4 rvo avoidance and you'll probably find some more info
they are two different systems which behave slightly differently from each other so you may want to try both to see which works better for you
oo thanks
@hallow flint when you have time could you give me a hand with the jumpscares? ๐ i have tried but im just really lost
Navigation Invokers is better or nav mesh ?
in performance
and i wanna use for realtime build
well navmesh are baked offline, so they should be faster
it entirely depends on your use case
sure, it might be later im at work
Thanks, although the UK isn't in the EU anymore, so we can basically turn into an AI slave market anytime.
offline ?!
what do u mean
navmeshes can be precomputed or dynamic
I think he means the former
god my head is like mush ๐
video meetings are... ๐คช
When you place down a nav mesh it does all the calculations to determine what is traversable or not, all before the game is running, offline
Need to make a deepfake of yourself paying attention to a meeting so you can stream that into the meeting while you do other stuff
yeah same lol
It was a bit crazy.. I've got a ton of follow-ups now
that's great, hopefully it leads to 
I'd say it was a success though, plenty of companies wanting to colab
Now we have to do some meetings to determine which partners we want to work with
And catch up with those we missed today
I think its fair to say, that lots of creative companies are interested in generative ML for their projects
Yeah, it was a bit crazy to be honest, like speed dating for VR companies ๐
one after another for 3 hours

well, 2 hours 45.. still, my head hurts ๐
Was this at the university or some sort of events where people look for opportunities to fund?
ahh i see, very cool. Will this fund the mocap lab you were mentioning?
apparently the UK MOD just got a 1.2bn boost ๐
naah, mocap lab is already funded
robotics dudes got funding for that
crazy
everyone getting a piece of the pie, maybe you'll get some of that too
MOD needs their sims
I'd settle for bus fare at this point ๐
LOL
They were all really impressed with my slide deck
which was basically just tons of AI generated images ๐
yeah, literally just spammed loads of DallE in there
it allows people to visualize
I used the images to complement the text on the slide.. so one sentense per slide, with at least 1 image per slide too
what kind of images you generate?
get on the MJ beta
Basically just random art that represented the points.. graphs, machines etc ๐
One I did today as a demo ๐
thanks for the heads up! just applied to beta
wow
that looks so good
Oh, hey Phil!
interesting, you can set the actual style with the last few words in the prompts
okay now how can i generate a 3d model from this haha
Yeah, prompt engineering is the new creative discipline
I saw some people on twitter posting analysis on why Dall-E Mini is going viral currently more than Dall-E 2.0
They were talking about all kinds of points
but... I mean... the only reason is that you can use Dall-E Mini right now
and you can't do that with Dall-E 2.0
:P
haha i was thinking the same thing
Well, yeah, you can access it freely, or run it locally
one is publicly available, one isn't haha
Better to use Disco Diffusion or something to be honest.. I don't think DallE-mini is a good example
DE2 is just so much better
oh btw.. DE2 generates tiling textures really nicely ๐
lemmie find an example
I was generating space goats for a fictional game jam game yesterday ๐ one sec..
I mean, in the realm of "shit I'd never be able to do" space goat art is up there..
Wow
There's also a certain 'trashy' quality to Mini that I think helps it go viral.
Sorry for spamming..
Mini is just messed up enough that it kind of undermines the concept ๐
Mini faces are Cronenbergian horrors ๐
Meeting, gotta go 
Hey all!
Apologize for a broad question, but I'm not sure where to start digging for best results.
I want to create a real-time "simulation", which involves significant numbers of AI-driven actors. The usual "live ecosystem" dream, you know.
And have it more of an honest sim, than smoke&mirrors approach usually taken in games.
So I have broad question, and a small specific one:
-
Any recommendations on proven approaches to handle large number of AI computations? Maybe well-known examples with documented info (like articles and GDC talks) to learn from? I'd assume games like Mount&Blade, or large-scale tactics/RTS games, have solved these issues at least partially.
-
If I want to include an option to fast-forward the time in Unreal, up to x10. Can that mess up the computations? If there are 500-1000 agents in the sim. Would it impact the performance significantly? Wondering if it's something I should think about in advance, when designing the architecture of the project and AIs.
Have you looked at the new Mass system from UE5?
there's a channel now.. #mass
also, check out the citysample projects on the UE marketplace
Nope, I'm clueless about that. I barely did some BTs in UE4, and that was all AI I did for many years ๐
In that case, go to the marketplace and download the citysample project
take a look ๐
I will research Mass, thanks for the headsup.
And also it now occurs to me that I've been coding in Blueprints only up until now, and for this project...it might now work, performance-wise ๐
for the time fast forward thing.. what do you want it to look like? does it have to be a sort of scrubbing thing like a video editor?
i.e. go back/forward in time
Oh check this out too: https://www.gdcvault.com/play/1022411/Massive-Crowd-on-Assassin-s
Yeah, ubi are basically the masters of this stuff.. done loads of GDC videos on it
Just a typical game speed-up. So the player can set it to fast speed and observe stuff changing fast, while still in real-time (no cuts).
Actually I don't need interaction to happen with the sim at high speed. Can be that you set pause, place stuff, unpause. Cheesy, but probably saves me from lot of headache.
there's a time dilation value you might be able to use for that
I think its "SetRelativeTimeDilation" or something
pretty sure the slowmo command uses that same system
Yeah, I think that was what I used on a city-builder project few years ago. I was wondering if it can break in edge cases like mine, with ton of AI computations happening
Well, you'd probably have to profile the high speed version to make sure the AI isn't killing your budget
But I don't think there's any real reason it wouldn't just work
mind you, haven't tried ๐
I might add that to my crowd assignment actually.. its a nice feature ๐
After discovering how values break at big enough distance from zero coord, I trust Unreal to break where I least expect it ๐ ๐
making my students create a crowd sim for drone training as their MSc graphics class ๐
This is unfortunately just a general thing that breaks in all game engines that handle coordinates as float values and most of them do - UE5 fixes it tho
they use doubles in UE5 now, so that shouldn't be the case anymore
Yeah, not intended as a stab at any particular engine, more like a commentary of "there are times where things we expect to just work, actually break"
well, they use a different numerical precision, I don't know if its actually doubles or some fixed point thing
yep
pretty sure it's doubles yeah
kind of makes sense, must have been a hellish change to test though ๐ฆ
All your unit tests would break ๐
Calling it now: after I read all of that material and look up examples, will probably end up going from hundreds of agents in the sim, to a couple dozen, half of which are always asleep anyway ๐
you have learnt the way of Unreal ๐
I think all my npc's still run CMC when they are dead bodies which is probably not great
Need to change it so it turns off CMC and just makes them physics props I think :p
good plan
Oh and here's an unreal engine implementation of this system
i don't have it but i've read the description and it seems to be based on this ubisoft system
here's a cheaper version of this same thing: https://www.unrealengine.com/marketplace/en-US/product/world-director-npc
Also i'm pretty sure there's some sort of limit in terms of how high you scan set TimeiDilation
i've tried increasing it before but unless my eyes tricked it, it seemed capped
and you'll definitely get issues with the physics
i think the physics engine might be your bottleneck? 
What if, hear me out.. I set up the default game speed at TimeDilation of 0.2 or so? Just with fast movement values on anything ๐ So that the x10 speed is not actually TD=10, but something like TD=2 ๐
Oh man, I haven't even started doing the thing, and already am designing crutches. This is gonna be fun.
loll does this not have to be seen at runtime?
It does. But it's abstract 3d figurines (like RimWorld, just in 3D static meshes..or like chess pawns, if you will), so...
ahh got ya
you could also make your own simplified movementComponent with some basic collision checks using geometry and linetraces. And you could accelerate your timedilation by a lot more
I'd say mass was the answer, depending on your representation and animation etc.
Yeah, I've noted it and the matrix city for research
I think a lot of it is going to depend on how much your sim can tolerate large timesteps, which will depend on how you write your update functions.
I have a gameclock implementation in my RTS learning project, and I definitely chased out bugs when I went from 1s/s to 1000s/s.
As I started to realize the scope and had reality checks, I've set up this sort of "challenge ladder" to climb:
- Plants AI. Conway would be proud.
- Simple animals (think Spore game, stage1).
- Social animals. Pack interactions, some behaviors like hunt-packs or task distribution within packs, etc.
And that was it, even at number 3 I felt like I'm already past comfort zone and will have plenty to learn and figure out. The scope and difficulty scales like crazy, heh.
But I want to learn more about how these things handles overall, top-down, before I start making sentient weeds in the bottoms-up approach.
Epic has a bit of code for a farm style plant growth AI using mass.. there's a stream discussing it and the code is in the engine, but apparently they haven't made an example to show it yet
Oh, don't have trouble with the 1-2, I did that in the past somehow. My problems start when it's about group interactions and scaling it into larger numbers / performance. Something I never had to worry about in past projects.
think someone did a mass sample for boids style movement.. check out their MassSample project on github
that's already so over my head. But also noted for future study, thanks!
This birdoids thing is awesome. Love it
has anybody successfully been able to interrupt an ai move task thats using goal tracking? I want my npc to adjust speed based on the distance of my character
It might be easier to just have the speed adjust while you're moving tbh
For example via using a service that checks distance and adjusts speed appropriately
I literally just got done implementing a service for adjusting the speed and it did work!
nice :)
Thanks! that does work perfectly!
no still at work 
like in 3 hours or so
hi guys
anyone know why my AI inst moving
isnt*
im using set paths
ive done the behaviour trees right
it just isnt moving
path points im doing
it wont "move to"
can anyone help
@hallow flint idk if you are any good with this, it just stays on โfindpathpointโ and doesnโt go on โmove toโ
if findpathpoint fails it would not go on to moveto since the parent is Sequence
probably why id guess
youd see this during runtime if you had the particular object set as Debug Object, there'll be a red indicator
It's also possibly you aren't calling FinishExecute in your FindPathPoint task? Tasks get stuck when that happens.
not exactly #gameplay-ai but is there a way to focus the player character pawn* in the gameplay debugger?
eh it probably wont help me since the gameplay debugger doesnt track changes to the controller anyway
click it on the world outliner
Anyone knows how can I debug AI is not moving to a location and after 5 secs pathfollowingcomp returns "Blocked"?
Whatever I do, it ends up on same result
Changed the target goal, changed the map, wrote custom move to node etc.
and there's definitely a green path when using navmesh visualization (P) ?
yep
gameplay debugger even shows the next path point to follow
its very close to AI
but its not moving
oh maybe check the min agent radius
possibly it is skipping it because the path isnt "wide" enough
ah, tbh I never edited agent settings since I started to UE. Lemme see what they do
omg i had no idea it did this, ๐ you da best
dude this is fantastic
I think both settings are fine ๐ค
(reads engine code) oh its the "stuck detection"
@celest python try calling SetMoveBlockDetection(false) on AIController and see if that fixes it
if so then i think its finding collisions nearby that block it
obviously youd want to find what is unexpectedly triggering stuck detection rather than remove it long term
never mind, stuck detection is purely "are we not moving when we thought we should"
Yeah, still wanted to give a luck but didnt work
so this actor never responds to AI movement or just cant move in a certain situation
At pathfollowingcomponent.cpp line 911, if (bHasNewSample && IsBlocked()) return true after 5 seconds and movement is aborted with blocked tag
until then, pathfollowingcomponent updates the path segments
and thinks its actually moving
but after that statement returns true if understands movement is blocked
sounds like the CMC configuration isnt right
root motion or direct?
obviously root motion has tons of gotchas you could be hitting
direct should be fine as long as CMC is set up right
nope, everything seems correct
well if a standard pawn works fine then hmm not sure
hmm, another pawn moves
ah ha!
so thats my fault rather than an engine error
thats some progress ๐ฅณ
thank you
probably one of the CMC settings or something else
let me know what it ends up being
always curious
alright, thanks again
it turns out my CMC subclass was corrupted in BP and I had to recreate it
hotreload strikes again
reinstancing 
I thought of a few solutions while using LogicDriver. Why people dont create array of structs/classes to define conditions and not have a single transition class? Think like each transition arrow/class has an array of instanced UObjects which checks particular things you define and allow transitions. Also communicating with previous state (the state its coming from) to allow transition evaluations or not. This way you also gain the ability of deciding if your current behavior in the state is interruptable or not.
From what I can see people always created a new transition class of each state they declared and it ended up being so complex and confusing. Unifying required classes and using new classes only to define new behaviors can let developers do their most of the job in the SM graph rather than coding new classes in IDE
Of course for very specific behaviors you end up creating new classes, or for scripted actions but wouldnt those ideas solve most of the things?
I think probably in the old times reflection systems were not that advanced so thats why people ended up creating classes everytime ๐ค
Hi i just seen this message
You still active @hallow flint ๐
I do have the finish execute
@hallow flint the blueprint class of the AI is a character
And the path point is an actor class
The orange part it stays on and the red part it doesnโt move to
can you show FindPathPoint
the coding for it?
yeah- assuming its custom, i didnt see it in the engine source in any case
oh honey
WAIT
if he didnt then hes not making a very good video
OH MY DAYS
THANK YOU
IT WORKS
sorta
it moves
but
it does the running animation instead of walking
any idea why
welp, check your Max Walk Speed on your Character Movement controller, and also check your Animation Blueprint to make sure you're handling that speed the way you want
what should the max walk speed on the ai contorller be
that depends on your animation blueprint ๐
and... you know
how fast you want it to move
if the speed feels about right, then take a look at your animation blueprint. Assuming you even have both a walk and a run animation, you probably would want to modify the blend space that controls whether walk or run is active
there is a zillion ways to do this stuff, im just guessing based on the usual tutorial path
the animation is fast or the movement speed* is fast
so the movement speed is fast
yes
then go to your character's movement component and adjust your Max Walk Speed
oki
also when it reaches the point, it stops for a split second and carrys on
so the animation stops
lol maybe
yes well it needs to rerun your other BT tasks and such
is there a way to add a "delay" between each movement to the point
ah yeah you can add delays in behavior tree
if you are trying to remove delays, it depends on what you are trying to do. A common use case is that the blackboard value is changing over time
but the moveto doesnt adjust on its own, it finishes what it started, and then does it again with the new value, but again, doesnt update live
in that case, use Observe Blackboard on the MoveTo
check out the Wait behavior tree node
yeah, technically anywhere under the sequence would probably do what you want
alright
remember, it executes them left to right
glad we were able to unblock you, always sucks to be stuck
wdum
wdym wdym, seems you were stuck for like... hours on this, now go forth and unreal ๐
OHH
ye
i was
thank you so much
now i gotta figure out the jumpscares ๐ฅฒ
this server is full of legends man
i was stuck for hours cuz i missed one line ๐ญ
it happens
i just need now for these 2 ai's to be completed before i can move on with the rest is the jumpscares, i need them to collide with the character or if its easier be in a certain distance with the player in order for a jumpscare animation to play
i can tell ๐ญ
hey @mint terrace if your still here, do you know how to do this? ๐
thats a little overbroad for a Q&A my friend
LOL
wdum ๐คฃ
is it possible to make it so when it reaches a curtain path point it changes to running instead of walking?
check 2nd rule on #rules
hi, does anyone know what's the proper way to setup a navmesh on a game with (sub)levels that are streamed/loaded at runtime? can't find much online and currently having a lot of problems
<@&213101288538374145>
what happened
was a spam bot there
ah
yo
can anyone help me out with the jumpscares? ive tried for a while and i just cant figure out what to do
i want it so when the ai collides with the player is plays the jumpscare animation
have you tried writing out in english what you want to happen?
what
usually, I recommend my students to write out the logic in comments first.. then write the code to do it
uh
basically, think through the logic of how you want it to happen
ok, so how do you improve to the point where you do know it?
Well, what if they don't?
ive tried youtube
you just going to not do something?
i am still searching yes
So try and do one part of it at a time
be methodical in your approach
again, write out the steps somewhere
and make sure you have a good understanding of whats going on at each step before moving on
one important thing, is to learn to debug and verify for yourself
ik ye
So I decided to learn how to write a Hierarchical Task Network and Slate at the same time, by jumping feet first into writing a graph editor to create an HTN asset.
Was this a good decision? Well, I'm learned a bunch and a) have a basic planner and b) a node editor has 60% of the right nouns in place. but on the other side, the way Epic writes UI's may drive me mad.
Also, Open pose is pretty neat. The model just barely *fits on my 2080, but with good lighting it's probably good enough for the simple mocap I wanted to do.
Hey folks
can create a child of ADetourCrowdAIController?
No, it's not exported. Just copy its functionality into your own AI controller, it's extremely simple
how?
I mean.. look at its code and just do the same thing in your own?
can u help me more xd
what is this
I mean you just need to copy the ObjectInitializer bits into your own controller
You can literally copy the code exactly one to one into a file in your own project
and just rename the class to something else
Yep
That should do it
all the detour controller does is it replaces the default path following component which is what manages pathfinding related movement with a different one
so it doesn't really need anything else
thanks
Ohhh you gonna open source it anywhere? I haven't dared touch Slate yet haha
I've been curious about how to make a graph editor, you referenced the BT code?
Once I have something another person can use, lol.
Haha I've only replicated the utility AI system I've seen in lectures
And I didn't even use it myself
Lol
Yeah. I can put it somewhere if you just want to see in progress mess, but yeah, I've used the BT editor, and https://github.com/jinyuliao/GenericGraph/tree/master/Source as references.
Getting a BB editor in looks rough: the existing one is only available by private include, so I am going to have to carefully build my own that doesn't copy the Epic one.
Yeah I'd be interested in checking it out
Is this for work or just for funsies
Just fun. I work in WebDev, this is my hobby programming outlet.
Ahh I see, yeah I do game dev for fun but I do occasionally get to do it work for as well lol
Lucky!
Someday I'll use this stuff in an internal prototyping project, but it's nice to have things to work on just to enjoy.
DM'd!
I'd recommend avoiding the BB stuff as an example.. its kind of terrible
and yeah +1 for doing your own graph editor.. its super fun
just make sure you understance MVC ๐
I think writing an editor is more difficult than writing a HTN planner ๐

