I love how it isn't perfect, like actual animals. The missed jumps were great. I very much "Oh, I missed" moment. xD And then when it stumbled and fell while running over the rough terrain- this is why you don't run your horses over uneven terrain~
The legs on a real dog extend further into a straight line after contact with the ground to provide a full spring, rather than remain curled after each contact with the ground.
german shepard's have weird back legs, and they often have hip problems, thats why the back legs seem scrunched, i think a doberman would have been a better reference.
You're Welcome the best reference must be a czekoslovakian or maybe a dingo or another basal race that would not have the genetic defects that most modern toy-races have.
0:19 walk 0:26 walk 1.2 m/s 0:26 trot 0:34 canter 0:39 Transverse gallop 1:32 Comparison with video footage. 3:46 Jumps 3:49 Jump across 3:53 Jump Down 3:56 Jump over 4:42 sitting 4:47 lie down
I was going to point that out too. It's a fantastic animation, it just jumps a little flat-footed (front and back feet hit the ground at the same time and an invisible string seems to be tugging the butt up level when the dog is propelling it'self into a leap. Looking at pictures of real dogs, the legs are usually shorter so they don't hunch when standing normally. Good model overall and amazing animal movements nonetheless. :D
I don't mean to be rude. Not at all, but I think during the jumps the shoulders on the front legs should definitely extend more, making the legs extend toward the neck.
I believe the major issue with the jumps is that they seem to be modeled after horizontal movement at high speeds, such as during hurdle races. At slow speeds, or jumping upward, animals tend to 'reach', rather than 'withdraw' (like airplane landing gears), their legs. This is because momentum carries you into the jump at higher speeds, so a jump consists mostly of applying force downward, then lifting legs clear of the obstacles.
Aaron this might be true, but these clips we're watching weren't animated by a person with prior knowledge. This is how the algorithm has evolved after 300 iterations. I imagine in 300 more, the motions will be more fluid, more complex, than they are now. It just hasn't figured that out yet.
@Akeiron85 Good eye... i think it's because the back legs are too long, and so they don't extend as they would (& do in the reference), and always seem pulled up, kind of like there is a joint problem or pain in the back legs limiting mobility.
Can someone PLEASE make a game about animals that utilizes this? I would love to play a game about cats or something that makes the cat /walk like a cat/, right down to the footfalls and balance. ovo
While I'll never have an interest in sports, you're definitely right. I like the games where you can't change direction, but maybe turn around? I can certainly turn around after a jump, but fully changing direction/momentum is nooot a thing.
Games let you do that because jumping is incredibly frustrating otherwise. Besides, you have way more control while in mid-air than you might think. This isn't obvious for something that happens quickly, like a jump, but it becomes very apparent when you are in the air longer, such as skydiving. Due to the combination of air resistance and changes in rotational inertia as you alter the pose of your body, you can substantially change the orientation, speed, and direction you are falling, which over a sufficient distance can have huge consequences. However, the average jump just doesn't have enough air time for this to matter much. then again, the same games that massively overdo your ability to change direction in a jump also involve jumps that are way beyond what's plausible. (To give a random example, from a stationary position mario can jump something like 5 times his own height, whereas a real person can maybe manage half their height.)
***** Sounds like an interesting game. And yes it is of course a bit of a historical artifact too. Mario can't suddenly have massively different movement physics. But a lot of platformers are also very cartoony. Realism isn't the goal for something like that anyway. On really old systems realistic physics was largely implausible of course. The hardware just wasn't powerful enough. But if you play old games you see quite a few that don't give you any control in the air whatsoever. They are incredibly frustrating to play, though it may also be related to the controls. (to have no control in the air means you need to be able to have a huge amount of control over the moment when you start the jump... Otherwise you effectively can't control where you land at all.)
you can try Kontrol ( google"puppet Kontrol") in browser, its a 2d universe sim game prototype with walking, swimming ragdolls if you want to see the playability. Timing and hoping is everything, lots of emergent gameplay.
@@yosha_ykt its at www.puppetarmyfaction.com you have to run iexplorer.exe , then try it in the old browser since Sivlerlight is now blocked everywhere. I have seen the stickman ragdoll fighting games now getting lots of success and I so plan to try to relaunch again in the App stores, string with Windowns and then doing Android.
@@yosha_ykt i just saw this notification. I generated you a full token if you want a free license to give feedback, if i do a new game it because mabye 2013 was too early for a physics walking game, enough people didn't appreciate it then. Recently I checked out Supreme Duelists and other Stick man games doing a similar method n weak gravity and using the legs to do about half the pushing walking, the other half being a magic force, and I realize people must get this now, maybe they are finally over Happy Wheels. Mine uses ZERO magic forces, and it lets you ride vehicles and swim. Most AAA games only use physics when you ide, and do not model forces of air even in flight simulations. My airship, bullets , everything is affected by winds in at least a simple way so you might croutch in strong winds to lower your profile. this is the link to th efull version , if you have suggestion put them to me, feedback is what shapes Kontrol. www.puppetarmyfaction.com/play.php?token=7fc9fd4c6654dea5be92c8ffd53bea1e
To be clear, there is no genetic algorithm being used. It does make some use of a derivative-free optimization algorithm known as CMA: "covariance matrix adaptation". The full paper can be found on the project page: www.cs.ubc.ca/~van/papers/2011-TOG-quadruped/index.html (as also included in the TH-cam description). -- cheers, Michiel
Hi, I would love to start working on a game based on this type of realistic physics and behavior. Could you tell me any books or information links where I can learn about all that stuff ? . Thanks a lot
Would have looked a lot more natural if you considered that quadrupedal animals lower their heads the faster they go. Especially when they sprint/chase something. It’s more aerodynamic and helps them to move faster.
Interesting, but the shoulder joint is way too high - almost humanoid- which makes the triceps muscle group less engaged in the stance phase of the foreleg. I have noticed this fact while my son is playing Elder Scrolls as a common mistake in his mounts (a skeletal wolf) in that the swing phase of the forelimb movement appears shortened. Just a note 7 years later....
I am reading through this paper, and I don't quite understand a few things, hope you can save me ;_; 1. For Fd(D), u describe that D measures forward progress in the gait and is computed as the ground-plane projection of Plf - Pfoot. Here the Plf means the origin (the square as shown in Fig.6) of the leg frame for the stance leg. However, in swing legs description, the target location P2 is calculated as Plf + (v - vd) * s. Here the Plf is described as the default stepping location relative to the leg frame for each leg which I cant quite understand what does it mean. 2. As the computation mentioned above: Plf + (v - vd) * s, what are v and vd? The velocity and the desired velocity of the swing foot? Sorry for my bad English, and thanks in advance.
ya sure it's simple. you just congregate the gait and rotary gallop to form the quantum-based Plf leg stance. then you add on the square root of the value of the Fd forward progress trot with the wackamended amount of dedicated wam and really to be totally honest I have no idea what I'm talking about you clearly know more than I do lmao XD
you where comparing it to the dog he had the leg sizes, the one on the video would probably be a lot more similar of the structure of a great dang since they tend to have longer legs. your justification for realism on the leg size seems to be contradictory unless you wanted to look like the same dog they showed.
Really nice but the movement seems a bit unnatural to me, his back legs should straighten more when running giving that "spring effect" which makes it so efficient, just my 2cents
WHAT THE FUCK. At the beggining I was listening "in the hall of the mountain king" but the volume never increased and I just founf out there was no sound at all.
Who's a good computer-generated boy? YOU'RE a good computer-generated boy!
coffeehauz lmfaaaooo
coffeehauz My thoughts exactly. I don't care if it's made of polygons, I still want to pet it. There's no hope for me.
*happy computerized buzz*
At last, we can finally have realistic dog AI for call of duty.
Eric V realistic and call of duty do not go together.. and really dude still playing cod? play a real game would you 😂
real game?
lmao he said real and game together
Clearly a simulation. The dog didn't circle about before laying down.
my dog does not walk in circles before laying down
you'r dog is clearly a simulation
LeChaosX
i knew my dog seemed too robotic :-(( it's all confirmed
I laughed way harder than I should have when it fell over at 2:40
I love how it isn't perfect, like actual animals. The missed jumps were great. I very much "Oh, I missed" moment. xD And then when it stumbled and fell while running over the rough terrain- this is why you don't run your horses over uneven terrain~
Great reference for quadruped movement and anatomical structure, thank you!
I read the title as "Locomotive Kills Simulated Quadrupeds"
I was confused for a while.
I love these type of videos.
I cant wait for the dog games of the future
This would be awesome for a platform/action/action-adventure game.
The legs on a real dog extend further into a straight line after contact with the ground to provide a full spring, rather than remain curled after each contact with the ground.
maybe it's not that perfect yet
how does it compare to cats then?
It doesn't want to get any more height than it absolutely needs to.
german shepard's have weird back legs, and they often have hip problems, thats why the back legs seem scrunched, i think a doberman would have been a better reference.
You're Welcome the best reference must be a czekoslovakian or maybe a dingo or another basal race that would not have the genetic defects that most modern toy-races have.
A even better reference is a tamaskan dog, they look like wolves and have no hip issues, their spine is a normal shape :)
0:19 walk
0:26 walk 1.2 m/s
0:26 trot
0:34 canter
0:39 Transverse gallop
1:32 Comparison with video footage.
3:46 Jumps
3:49 Jump across
3:53 Jump Down
3:56 Jump over
4:42 sitting
4:47 lie down
Andrej Karpathy is clearly an all pervasive being. It's like he's everywhere!
I was going to point that out too. It's a fantastic animation, it just jumps a little flat-footed (front and back feet hit the ground at the same time and an invisible string seems to be tugging the butt up level when the dog is propelling it'self into a leap. Looking at pictures of real dogs, the legs are usually shorter so they don't hunch when standing normally. Good model overall and amazing animal movements nonetheless. :D
This is so awesome
is this how to beat qwop
This is how evolution beats QWOP
There is a video like this for qwop check deep learning qwop or something like that.
Or Clop
please please please, release this program? Even if it was payed, I'd buy it, would be really useful.
That was really cool. Liked watching the first examples the best. The whole video is great though.
what software do I need to build these models?
Good boy simulator 2011
what is this software
can we use it for out own work are there any way doing this kind of simulation.
I don't mean to be rude. Not at all, but I think during the jumps the shoulders on the front legs should definitely extend more, making the legs extend toward the neck.
I believe the major issue with the jumps is that they seem to be modeled after horizontal movement at high speeds, such as during hurdle races. At slow speeds, or jumping upward, animals tend to 'reach', rather than 'withdraw' (like airplane landing gears), their legs. This is because momentum carries you into the jump at higher speeds, so a jump consists mostly of applying force downward, then lifting legs clear of the obstacles.
Aaron this might be true, but these clips we're watching weren't animated by a person with prior knowledge. This is how the algorithm has evolved after 300 iterations. I imagine in 300 more, the motions will be more fluid, more complex, than they are now. It just hasn't figured that out yet.
Who are you telling to improve
this is ridicoulously helpful
Is there any plans to make this a commercial product with FBX exports for crowd sims? or even arthropods?
@Akeiron85 Good eye... i think it's because the back legs are too long, and so they don't extend as they would (& do in the reference), and always seem pulled up, kind of like there is a joint problem or pain in the back legs limiting mobility.
looks so real
That 5.0m/s transverse gallop on diverse terrain one CERTAINLY died.
2:40 lol ded
it is really good, what is the basic logic about the dog？ how does it keep its balance? great.
That was so awesome! :-D
Captions: leap over 2m gap
Dog: bonk
Captions: let's try that again shall we
1:52 pupper do a heckin _transverse gallop_
2:39 R.I.P DIGI-DOGGO 2011-2011
You won't be able to gotta go fast anymore 😪
Can someone PLEASE make a game about animals that utilizes this? I would love to play a game about cats or something that makes the cat /walk like a cat/, right down to the footfalls and balance. ovo
Careful, you don't want cats too close to you, Sandbox.
Pfff, I have a cat sleeping on my chest every night, they aren't *that* scary X0
While I'll never have an interest in sports, you're definitely right. I like the games where you can't change direction, but maybe turn around? I can certainly turn around after a jump, but fully changing direction/momentum is nooot a thing.
Games let you do that because jumping is incredibly frustrating otherwise.
Besides, you have way more control while in mid-air than you might think. This isn't obvious for something that happens quickly, like a jump, but it becomes very apparent when you are in the air longer, such as skydiving.
Due to the combination of air resistance and changes in rotational inertia as you alter the pose of your body, you can substantially change the orientation, speed, and direction you are falling, which over a sufficient distance can have huge consequences.
However, the average jump just doesn't have enough air time for this to matter much.
then again, the same games that massively overdo your ability to change direction in a jump also involve jumps that are way beyond what's plausible.
(To give a random example, from a stationary position mario can jump something like 5 times his own height, whereas a real person can maybe manage half their height.)
***** Sounds like an interesting game.
And yes it is of course a bit of a historical artifact too.
Mario can't suddenly have massively different movement physics.
But a lot of platformers are also very cartoony. Realism isn't the goal for something like that anyway.
On really old systems realistic physics was largely implausible of course. The hardware just wasn't powerful enough.
But if you play old games you see quite a few that don't give you any control in the air whatsoever.
They are incredibly frustrating to play, though it may also be related to the controls.
(to have no control in the air means you need to be able to have a huge amount of control over the moment when you start the jump...
Otherwise you effectively can't control where you land at all.)
the ultimate spore 2 engine
make this a simple game. slap it on the ios store. profit
you can try Kontrol ( google"puppet Kontrol") in browser, its a 2d universe sim game prototype with walking, swimming ragdolls if you want to see the playability. Timing and hoping is everything, lots of emergent gameplay.
@@DamianHallbauer give me a link please
@@yosha_ykt its at www.puppetarmyfaction.com you have to run iexplorer.exe , then try it in the old browser since Sivlerlight is now blocked everywhere. I have seen the stickman ragdoll fighting games now getting lots of success and I so plan to try to relaunch again in the App stores, string with Windowns and then doing Android.
@@yosha_ykt i just saw this notification. I generated you a full token if you want a free license to give feedback, if i do a new game it because mabye 2013 was too early for a physics walking game, enough people didn't appreciate it then. Recently I checked out Supreme Duelists and other Stick man games doing a similar method n weak gravity and using the legs to do about half the pushing walking, the other half being a magic force, and I realize people must get this now, maybe they are finally over Happy Wheels. Mine uses ZERO magic forces, and it lets you ride vehicles and swim. Most AAA games only use physics when you ide, and do not model forces of air even in flight simulations. My airship, bullets , everything is affected by winds in at least a simple way so you might croutch in strong winds to lower your profile. this is the link to th efull version , if you have suggestion put them to me, feedback is what shapes Kontrol. www.puppetarmyfaction.com/play.php?token=7fc9fd4c6654dea5be92c8ffd53bea1e
Is there a link, a way, to find these animation files as BVH ?... This would be really helpful !!!
Thx.
What programme do you use to run the genetic algorithm?
To be clear, there is no genetic algorithm being used. It does make some use of a derivative-free optimization algorithm known as CMA: "covariance matrix adaptation". The full paper can be found on the project page: www.cs.ubc.ca/~van/papers/2011-TOG-quadruped/index.html
(as also included in the TH-cam description). -- cheers, Michiel
Hi, I would love to start working on a game based on this type of realistic physics and behavior. Could you tell me any books or information links where I can learn about all that stuff ? . Thanks a lot
+PianoLife have you found anything?
no
Would have looked a lot more natural if you considered that quadrupedal animals lower their heads the faster they go. Especially when they sprint/chase something. It’s more aerodynamic and helps them to move faster.
I want that program.
I want to play around with it.
Ok I could not tell the difference between trot and pace
Interesting, but the shoulder joint is way too high - almost humanoid- which makes the triceps muscle group less engaged in the stance phase of the foreleg. I have noticed this fact while my son is playing Elder Scrolls as a common mistake in his mounts (a skeletal wolf) in that the swing phase of the forelimb movement appears shortened. Just a note 7 years later....
what program is that? :)
this is satisfying
Any cuber here?
If yes, BADMEPHISTO worked on this.
Andrej Karpathy.
I just tried it. It's great. It would be awesome if the animation can be recorded and exported to FBX for use in MotionBuilder Quads. Thanks.
for god's sake stop torturing the damn rig...
great reference, thanx!
When you work in a graphics company and gamble too much on horses , then you get this video.
SO INTERESTING
I wish there were more details on how it works though
+David M. Johnston did you click the link in description?
Tom H I know I should but they could have given a bit of explanation in the video already. I'll check that out latter perhaps.
I am reading through this paper, and I don't quite understand a few things, hope you can save me ;_;
1. For Fd(D), u describe that D measures forward progress in the gait and is computed as the ground-plane projection of Plf - Pfoot. Here the Plf means the origin (the square as shown in Fig.6) of the leg frame for the stance leg.
However, in swing legs description, the target location P2 is calculated as Plf + (v - vd) * s. Here the Plf is described as the default stepping location relative to the leg frame for each leg which I cant quite understand what does it mean.
2. As the computation mentioned above: Plf + (v - vd) * s, what are v and vd? The velocity and the desired velocity of the swing foot?
Sorry for my bad English, and thanks in advance.
ya sure it's simple. you just congregate the gait and rotary gallop to form the quantum-based Plf leg stance. then you add on the square root of the value of the Fd forward progress trot with the wackamended amount of dedicated wam and really to be totally honest I have no idea what I'm talking about you clearly know more than I do lmao XD
俞達
Yes, v is velocity and vd is desired velocity.
Can't help with the first question as it is above my intelligence level, sorry
2:40 wait so the dog fell over irl? lol
isnt dis video available in slow motion
?
Kk
So an AI did this? A computer? Made a perfectly simulated realistic dog?
This is made using a genetic network algorithm.
4:26 Best part.
so how i can get it for maya ?
You could use something like this to make the ultimate QWOP player.
you where comparing it to the dog he had the leg sizes, the one on the video would probably be a lot more similar of the structure of a great dang since they tend to have longer legs. your justification for realism on the leg size seems to be contradictory unless you wanted to look like the same dog they showed.
It\s bad animation. What happened with back legs? Where overlap?
Семён Шитиков это не анимация. Это физическая симуляция. Все происходит на лету
Она какая-то странная, правда. Многих движений нет, если это мокап, ощущение, что он что-то не снял или его плохо почистили. Желаю удачи.
Wait so what is this?
Like a model learning how to walk and stuff? Haha I'm not into computers and all that but this looks really cool.
good
"Leap over 2 meter gap while trotting, take 1"
AWWWW i wanted to hug the poor boi =[
now, use that skill to put an AI in this good boy, show it to a company and you're set for life
2:40 WTF happened to that dog?
is he ok?
Im an animator but this is nice!
would work wonders as a replacement to animation in games, could just simulate these structures in real time instead.
i swear The Last Guardian used this, it'd be insane if they didn't
Really nice but the movement seems a bit unnatural to me, his back legs should straighten more when running giving that "spring effect" which makes it so efficient, just my 2cents
I think it's a bit weird due to lack of weight in the tail, I mean, dogs use it more to be more stable
2:40 NOOooOoooOoOoOOOOOOooo
Doggo
try doing this but with a tripod
my god ! i.... i... jizzzzzed in my pants
Everything looks great till the jumps those look terrible like the dog doesn’t even try to land with its front legs
Look like few years of dog simulation.
Is nobody going to talk about how he almost broke his neck in 4:23
WHAT THE FUCK.
At the beggining I was listening "in the hall of the mountain king" but the volume never increased and I just founf out there was no sound at all.
TIRICER!
-SWANGIN-
2:40 oof
Having at some sound would have been nice. At least some elevator music
2:40 lmfao! poor thing!
Аааа как я тут оказался?
ROBOT DOGS PEOPLE
I think i broke my dog
Is there any plans to make this a commercial product with FBX exports for crowd sims? or even arthropods?