Fast talk or not - Big respect for her work - and everyone can be nervous. The talks are delivered by real people, because they have real things to say.
What's really interesting is that if you turn on the auto-generated captions in this video, they exactly match what she is saying. The algorithm is picking up on her words with way more accuracy than most speakers. Just listen to her articulation. Its perfect.
Super impressed that she managed to present so much good info in the tiny scrap of time GDC gave her. Half hour talks are just a bad idea. The audience are developers. They want depth, not summaries. But anyhow, she mentioned constraint solving with "IK" solvers. I assume this is something different from Inverse Kinematics used for animation? Not having much luck searching around to figure out what it stands for so I can find more info. I had to laugh when she followed up mentioning it with 'don't write your own!' as I'd already opened a tab on another monitor to look into just that...
Yes. Inverse Kinematics. You had it right. Its a way to allow a sort of programming for joints so they can interact with the environment and be reactive instead of just an animation playing upon command, regardless of environment.
Inverse Kinematics wasnt all that hard the last time I tried, but that was under the assumption that approximations were sufficient, and you had the knowledge of how to do it well. If you dont have the academic knowledge of how to solve these kinds of problems, then yes it would probably be a poor idea to write them yourself. I have also worked with some other kinds of constraint problems, and while they might not be beginner problems, there are a lot of other things I would be far more hesitant to write myself. A simple feeed foward neural network requires more work in implementing than a simple constraint solver, at least the ones I worked with, while a support vector machine being something I would not want to completly implement by myself (though that does require one to solve several contraint problems). If one is familiar with Lagrange multipliers and/or some of the other techniques (depending on what type of problem to solve), then it becomes quite feasable to do. That said, I would not throw this at some random code monkey, as they would be stuck, but if you know what you are doing, then it isnt all that bad.
I think this might be my favorite GDC talk of all time. I comment just now, but I've watched this video multiple times and apply some of the principles Kate discusses here on a day to day basis. Really great talk. A brilliant mind and excellent communicator!
It's OK. I've watched Gilmore Girls. This talk was PHENOMENAL. This is one of those talks that I have in stock all day, waiting for a break to see it, and then it turns out to be so amazing and invigorating, that it fully recharges my batteries and jolts me back to work!
This talk was super interesting. I feel like a lot of the points that re just briefly mentioned could be entire talks on their own, so it's unfortunate that she didn't have more time.
Great talk, even greater respect for her work. For me the 10 000 oat meals problem is a major highlight of this talk, because so many games suffer from it when it comes to generating terrain. For me it could be summarized as an ocean of content with the depth of a puddle. Take Minecraft, infinite world, countless amount of them, with many different biomes. But in most cases once you've seen a biome, you've pretty much seen any possible variation of it. No Mans Sky, you have an infinite amount of solar systems and planets. Sure, every planet is different from the last one, but once you've landed on the surface and spent more than five minutes on it, you've pretty much seen the entire planet, as there are no biomes.
Outstanding overview, exactly as described. I had no trouble following and understanding the presentation. Excellent use of description and visual so we could see what was being discussed. Each approach described could obviously bear its own hour(s) long presentation, which I would absolutely watch (assuming it is not already out there, will check). Thank you!
Procedural generation difficulty varies based on the desired outcome. Example: Procedurally generating infinite, immutable terrain is easy mode, whereas procedurally generating a full-size planet with LoD scaling terrain you can land on is quite a bit more difficult. Procedurally generating life forms ala No Man's Sky is roughly in between the two. I've been researching the topic for several years now.. It's fascinating stuff!
What a densely packed Talk...! The Blog post, this video & the Resources mentioned in the last slide are going to keep me busy for a while.... Super Informative lecture by a very Talented Artist.... Her work really is awe-inspiring ! PS: I just googled Endless Screaming twitter bot :P .... you learn something new everyday !
To one of the questions about performance considerations on results: It's important to visualize this sorta stuff as a n-dimensional vector. It's a data container. If it isn't performant, but you have some other thing that is, you can interpolate between those two positions to achieve a compromise. It's what your GANs are doing when they are testing outcomes. You can do all sorts of curve-fitting on data to make it conform to constraints. You just have to be clear where point A is and where point B is. From there, it's just math.
This talk was way better than I was expecting, honestly. 25:00 I think she mentions No Man's Sky here, but it's also a shot across the bow to Borderlands and their... what, 4 billion different guns?
I apologize for the very specific question but; How was the River Dredging generated when building worlds in Sore? Thank you for such a well made presentation!
If you want access to the catalog she mentioned, you can find it on her itch. I can't be more specific because my comment saying this keeps getting deleted, by some bot I assume.
I'm frustrated that this talk had to be so short, she can't even pace the information comfortably. Not a diss on her though, I'm super interested in the topic, it deserves an hour and a half
Very useful tutorial. I need to know which algorithm is best suited for creating a building in unreal engine with given vertices and height in the real time coordinates
Aw man, real cool to see what looks like Strangethink's Secret Habitat mentioned and shown. I have a copy of that somewhere in my machine; neat little game.
does anyone know where to find the game she shows in the slide at 24:37? I remember playing around with it and I wanted to look it up again , but I forget how to type it in. Very fun interactive three panel narrative
A good idea of augmented reality, would be a sattelite with a quantum supercomputer, thst projects a digital signal to a smartphone, that would procedurally generate.... tutorial tips for your day to day life. Like recipes, or paths to items you have on your grocery list Or a list of the medicinal or other qualities of a given plant encountered in the wild or otherwise.
I think Ashes of Creation's node system is begging to be tweaked and converted to use tiles (9:44). Anyone know what Intrepid Studios is going to try to randomly generate?
Too fast. The monotone rush makes it too easy to zone out while watching. I can't help but feel the speech would have been better if it had been better edited for time, rather than rushed.
What if you just love PCG, want to try to make the best procedure world generator possible -- and then, since you have to the world, perhaps go ahead and build a game around it?
I mean, if I already didn't know most of these things, most of the time I wouldn't have understood what exactly she means or how exactly is this technique used. Such a shame she wasn't given more time.
Do you know what is the best part of it? Some members of the audience went home and cried alittle , and they came back to comment this video like they actually understood something. Come on guys , "for everyone" was the joke in it, calm down :D ...
The problem with simulation is that the world is complex and inefficient to simulate. Also, pre-simulation is incompatible with infinite generation of a continuous simulated area. For example, how *not* to make a Spore clone: For every planet, begin by taking the absolute value of a noise map to divide it into continents. 2. simulate plate tectonics, continental drift, and seafloor spreading, as well as erosion and volcanism as you seed the planet with microbes that have randomized brains and bodies. 3. Simulate the microbes moving around, eating, breeding, mutating, etc. 4. Once a species develops to a certain size, allow all species to begin migrating to land. 5. Once the migration to land is complete, begin mutating and breeding the simulated land creatures until either the planet is as old as it should be or one of the creatures learns to use tools. 6. Then freeze biological evolution and plate tectonics and simulate tribal living until a continent has unified or the time limit is reached. 7. Then simulate the civilization scale strategic decisions and vehicle combat of warring empires until either the time has elapsed or one empire conquers, converts, or buys out the rest. Finally, simulate any imperial expansion into space, chance of rebellions, ideology drift, trade, etc. The problem with this is that while it could produce convincing results, it is far, far too slow to simulate all of this when you just show up at a planet at random and the game suddenly needs to decide what's there. You need to be able to do that in less than a second.
re: "10000 bowls of oatmeal problem" Yep. How much difference is there between one 20x20 km area in a Minecraft world and another 20x20 km area? Statistically speaking they're nearly the same. You may have 9 million times that area before you reach the world border but it's very repetitive past a certain distance. Not to mention every seed that isn't broken is functionally identical on that scale.
Problem with procedural generation. Is that whilst it can be extremely scalable and flexible. It inherently ends up having far less detail then hand crafted terrain or maps or creatures or whatever. And generally you wont end up playing a game long enough to justify generating for example in No Man's sky 4 billion or whatever planets. Though i'm not bashing procedural generation, it is defiantly useful when used correctly. For example in Star Citizen they are using procedural generation for certain parts of planets such as terrain and biosphere within certain parameters based off what type of planet they want to make. However they always have the ability after generation to implement detail in order to make the planets feel far more alive. So it pretty much comes down to generation for scale, however it's important to have hand crafted detail.
Fast talk or not - Big respect for her work - and everyone can be nervous. The talks are delivered by real people, because they have real things to say.
unlike all theese internet bitches in their little troll shacks, void of light and social interaction and showering for month.
I'd kick your ass if I wasn't 350 lbs and could get off my rascal scooter.
Great talk.
TBH she doesn't sound nervous, just short on time.
What's really interesting is that if you turn on the auto-generated captions in this video, they exactly match what she is saying. The algorithm is picking up on her words with way more accuracy than most speakers. Just listen to her articulation. Its perfect.
"talk slowly" my teachers said, and yet this feels like the perfect pace for a talk
I keep coming back to this video, amazing and useful ideas in here, Kate Compton is truly one of the greats.
I have to keep stopping this video every 2 minutes to google the all the references she's giving out. This is excellent!
This was really good. She needs to come back and get a reasonable time slot.
Yeah! She is amazing. Hope to see her soon. Sad that it took me years to find her
Super impressed that she managed to present so much good info in the tiny scrap of time GDC gave her. Half hour talks are just a bad idea. The audience are developers. They want depth, not summaries. But anyhow, she mentioned constraint solving with "IK" solvers. I assume this is something different from Inverse Kinematics used for animation? Not having much luck searching around to figure out what it stands for so I can find more info. I had to laugh when she followed up mentioning it with 'don't write your own!' as I'd already opened a tab on another monitor to look into just that...
Yes. Inverse Kinematics. You had it right. Its a way to allow a sort of programming for joints so they can interact with the environment and be reactive instead of just an animation playing upon command, regardless of environment.
Inverse Kinematics wasnt all that hard the last time I tried, but that was under the assumption that approximations were sufficient, and you had the knowledge of how to do it well. If you dont have the academic knowledge of how to solve these kinds of problems, then yes it would probably be a poor idea to write them yourself.
I have also worked with some other kinds of constraint problems, and while they might not be beginner problems, there are a lot of other things I would be far more hesitant to write myself. A simple feeed foward neural network requires more work in implementing than a simple constraint solver, at least the ones I worked with, while a support vector machine being something I would not want to completly implement by myself (though that does require one to solve several contraint problems).
If one is familiar with Lagrange multipliers and/or some of the other techniques (depending on what type of problem to solve), then it becomes quite feasable to do. That said, I would not throw this at some random code monkey, as they would be stuck, but if you know what you are doing, then it isnt all that bad.
@@kris3451 thank you very much. I had trouble understanding what she meant by brute forcing but you explain very well.
@John Smith "Rescue on fractulus" Was the first procedural game I saw (Atari 130xe version, so mid 1980s here), that was really ahead of its time.
Next year, can Kate have more time? Like, lots more? but still talk at the same speed?
Nicholas Bieber I think she should be given one hour but should cover like one topic in depth.
Maybe we can agree on, like, 5% slower
Why was she only given 30 minutes? >.< Excellent talk on a really cool subject, though.
@@AllahDoesNotExist Fuck your hateful comment if I can't respond to it in kind.
I think this might be my favorite GDC talk of all time.
I comment just now, but I've watched this video multiple times and apply some of the principles Kate discusses here on a day to day basis.
Really great talk. A brilliant mind and excellent communicator!
this is the best talk about PGC. Too short, very interesting, well explained, great experience she has.
Searching more video of Kate Compton on UA-cam was not exactly what I expected
second talk I've seen by Kate on procgen, she's amazing, huge expertise here. Extremely grateful to hear her wisdom
These speed-download talks are my favorite! no fluff, just data!
It's OK. I've watched Gilmore Girls.
This talk was PHENOMENAL. This is one of those talks that I have in stock all day, waiting for a break to see it, and then it turns out to be so amazing and invigorating, that it fully recharges my batteries and jolts me back to work!
Wow this presentation is fantastic. Big thanks to Kate for sharing only a fraction of her experience. Definitely getting into PCG!
I've watched so many gdc talks and this is easily one of the best.
She was great to listen to, it's clear she is passionate about it and likes to share her experiences with us
This talk was super interesting. I feel like a lot of the points that re just briefly mentioned could be entire talks on their own, so it's unfortunate that she didn't have more time.
Great talk, even greater respect for her work.
For me the 10 000 oat meals problem is a major highlight of this talk, because so many games suffer from it when it comes to generating terrain. For me it could be summarized as an ocean of content with the depth of a puddle.
Take Minecraft, infinite world, countless amount of them, with many different biomes. But in most cases once you've seen a biome, you've pretty much seen any possible variation of it.
No Mans Sky, you have an infinite amount of solar systems and planets. Sure, every planet is different from the last one, but once you've landed on the surface and spent more than five minutes on it, you've pretty much seen the entire planet, as there are no biomes.
Outstanding overview, exactly as described. I had no trouble following and understanding the presentation. Excellent use of description and visual so we could see what was being discussed. Each approach described could obviously bear its own hour(s) long presentation, which I would absolutely watch (assuming it is not already out there, will check). Thank you!
Procedural generation difficulty varies based on the desired outcome. Example: Procedurally generating infinite, immutable terrain is easy mode, whereas procedurally generating a full-size planet with LoD scaling terrain you can land on is quite a bit more difficult. Procedurally generating life forms ala No Man's Sky is roughly in between the two.
I've been researching the topic for several years now.. It's fascinating stuff!
"A space of space creatures" - Brilliant.
Really love this talk, and all of the great details! Amazing work, Kate.
Fantastic talk. Thank you so much
I talk really when am in my normal state, she is doing great honestly, i loved her way of explaining stuff. Absolutely exciting
If she didn't get a standing ovation, I swear to God...
Great to hear from such an expert on a subject.
those gdc videos need some subtitles, for people who can read in english but are weak in hearing.
Yeah men use the auto cc, or maybe you can post your own cc
CC is under the video on the right.
Automatic CC wasn't enabled at the moment it launched.
Damn, my friend, better grammar skills than most, probably even me!
What a densely packed Talk...! The Blog post, this video & the Resources mentioned in the last slide are going to keep me busy for a while.... Super Informative lecture by a very Talented Artist.... Her work really is awe-inspiring !
PS: I just googled Endless Screaming twitter bot :P .... you learn something new everyday !
To one of the questions about performance considerations on results:
It's important to visualize this sorta stuff as a n-dimensional vector. It's a data container. If it isn't performant, but you have some other thing that is, you can interpolate between those two positions to achieve a compromise. It's what your GANs are doing when they are testing outcomes. You can do all sorts of curve-fitting on data to make it conform to constraints. You just have to be clear where point A is and where point B is. From there, it's just math.
She was so inspiring to listen to.
She is an absolute treasure
An angry nun in a wedding dress?! Did that fanfic ever come to life? I am intrigued!
I speed up the audio usually anyway, so I appreciate her speech cadence. gogogogo.
Now here is someone who is passionate about what they do...
This is a good entry point video to PCG, since she provide a good overview a lot of follow ups and references.
Great presentation, Kate. Jam packed with great stuff.
awesome talk! Very informative and loved the speed
Amazing talk! Very impressive! :)
my thirst for the information and her verborrheal delivery is a perfect match, could've watched two hours of her just explaining things
best PG talk on this channel yet!
Loved this. Real info. Learned lots. A+
Watched this at 2x speed, great talk! She needed more time )
This talk was way better than I was expecting, honestly.
25:00 I think she mentions No Man's Sky here, but it's also a shot across the bow to Borderlands and their... what, 4 billion different guns?
Loved this talk , amazing woman!
What an inspiring programmer!
Damm seeing love for dwarf fortress from 2017 in retrospect is amazing now that the game is released on steam finally and a big hit
Pretty good talk. Wish she had more than a half hour!
Excellent speech!
This is such an amazing talk
I apologize for the very specific question but;
How was the River Dredging generated when building worlds in Sore?
Thank you for such a well made presentation!
If you want access to the catalog she mentioned, you can find it on her itch. I can't be more specific because my comment saying this keeps getting deleted, by some bot I assume.
Can we get a procedurally generated keynote speech?
Great presentation!
I'm frustrated that this talk had to be so short, she can't even pace the information comfortably. Not a diss on her though, I'm super interested in the topic, it deserves an hour and a half
Very useful tutorial. I need to know which algorithm is best suited for creating a building in unreal engine with given vertices and height in the real time coordinates
Great talk, thanks Kate!
My work blocks URL shorteners. Can someone post the full URL to the slides?
Aw man, real cool to see what looks like Strangethink's Secret Habitat mentioned and shown. I have a copy of that somewhere in my machine; neat little game.
where can I find the blog post?
She's awesome, and the action rogue-like game she mentioned called Unexplored is also great. Can't believe someone else has even heard of it.
You had me at "I'm good at counting"
does anyone know where to find the game she shows in the slide at 24:37? I remember playing around with it and I wanted to look it up again , but I forget how to type it in. Very fun interactive three panel narrative
Amazing talk, She should really go in depth for her next talk. Would love to watch that
Just realized this video is exactly a year old
A good idea of augmented reality, would be a sattelite with a quantum supercomputer, thst projects a digital signal to a smartphone, that would procedurally generate.... tutorial tips for your day to day life. Like recipes, or paths to items you have on your grocery list
Or a list of the medicinal or other qualities of a given plant encountered in the wild or otherwise.
Great content! You get used to the pace in a few minutes.
Talk begins at 6:08
Deep Respect - great talk!
Great talk, that was impressive, I got some inspiration for my Blender Sushi vlog.
Awesome talk!
Someone knows where i can find the wallpaper at 22:52?
I think Ashes of Creation's node system is begging to be tweaked and converted to use tiles (9:44). Anyone know what Intrepid Studios is going to try to randomly generate?
thanks for this intro! so good!!!!!!!
I love this kind of stuff
OMG!!!... tnx!! this is awesome!!!
she is so awesome!
Woah thats some fast talk :D
Too fast. The monotone rush makes it too easy to zone out while watching. I can't help but feel the speech would have been better if it had been better edited for time, rather than rushed.
not fast enough? try x1.5 speed
lol I watched it at 2x speed.
If you slow it down it's easier to grok but it makes her sound drunk 😄
Procedurally generate your preferred speed by going to the settings and slowing down the audio.
For nine of her 30 minutes she talks about what she is going to talk about - content generation at the best ;)
What if you just love PCG, want to try to make the best procedure world generator possible -- and then, since you have to the world, perhaps go ahead and build a game around it?
24:44 Do you have the source for that murder mystery generator? Would definitely want!
her built in words generator is on overload mode
I mean, if I already didn't know most of these things, most of the time I wouldn't have understood what exactly she means or how exactly is this technique used. Such a shame she wasn't given more time.
She convinced me in Spore
At 0.85 speed this talk sounds about right
I would like for there to be a proceerually generated choose your own adventure story.
AI Dungeon!
Anybody have a link to the "zines" mentioned at the start of the video?
Fnatastic content !!! Thank you !!!
Do you know what is the best part of it? Some members of the audience went home and cried alittle , and they came back to comment this video like they actually understood something. Come on guys , "for everyone" was the joke in it, calm down :D ...
Pro tip: set it to 0.75 speed
Spore galactic adventure won't start on my PC, anybody know how to fix it? I bought it off of origin.
6:52 what did she say her tool was?
Alex Tracery
19:12
Me: Out of breath
Her: I'm already behind !
Jokes aside, really cool talk.
Bravo!
fun ^ 10 x int ^ 40 = Ir2? 23:26
"The spore creature creator was notoriously really really good."
If only the combat, social, and progression system were as good. 😭
The problem with simulation is that the world is complex and inefficient to simulate. Also, pre-simulation is incompatible with infinite generation of a continuous simulated area.
For example, how *not* to make a Spore clone:
For every planet, begin by taking the absolute value of a noise map to divide it into continents.
2. simulate plate tectonics, continental drift, and seafloor spreading, as well as erosion and volcanism as you seed the planet with microbes that have randomized brains and bodies.
3. Simulate the microbes moving around, eating, breeding, mutating, etc.
4. Once a species develops to a certain size, allow all species to begin migrating to land.
5. Once the migration to land is complete, begin mutating and breeding the simulated land creatures until either the planet is as old as it should be or one of the creatures learns to use tools.
6. Then freeze biological evolution and plate tectonics and simulate tribal living until a continent has unified or the time limit is reached.
7. Then simulate the civilization scale strategic decisions and vehicle combat of warring empires until either the time has elapsed or one empire conquers, converts, or buys out the rest. Finally, simulate any imperial expansion into space, chance of rebellions, ideology drift, trade, etc.
The problem with this is that while it could produce convincing results, it is far, far too slow to simulate all of this when you just show up at a planet at random and the game suddenly needs to decide what's there. You need to be able to do that in less than a second.
you have described Dwarf Fortress
Good talk
superb work. I paused ever slide because of her verbal insights.
my typing on an ipad force reference.
re: "10000 bowls of oatmeal problem"
Yep. How much difference is there between one 20x20 km area in a Minecraft world and another 20x20 km area? Statistically speaking they're nearly the same. You may have 9 million times that area before you reach the world border but it's very repetitive past a certain distance. Not to mention every seed that isn't broken is functionally identical on that scale.
Problem with procedural generation. Is that whilst it can be extremely scalable and flexible. It inherently ends up having far less detail then hand crafted terrain or maps or creatures or whatever. And generally you wont end up playing a game long enough to justify generating for example in No Man's sky 4 billion or whatever planets. Though i'm not bashing procedural generation, it is defiantly useful when used correctly. For example in Star Citizen they are using procedural generation for certain parts of planets such as terrain and biosphere within certain parameters based off what type of planet they want to make. However they always have the ability after generation to implement detail in order to make the planets feel far more alive. So it pretty much comes down to generation for scale, however it's important to have hand crafted detail.