Machine Learning Summit: Ragdoll Motion Matching

[@lonelydoraiba]

I cant stop watching 1:57 - 2:24 this is incredible to see, let alone, to be able to do

[@Build_the_Future]

I'm building a 2 legged full-size robot using Motion Matching on my channel. Check it out and tell me what you think. Thanks

[@vast634]

In GTA 4 they already used Euphoria Physics, wich is very similar, making bipedal actor dynamically react to their pose.

[@khangvinh4656]

@@vast634 same, but the problem with Euphoria is it being extremely hard to implement while this version is aiming to improve that process making it more accessible to the public🤘

[@anymyo9882]

@@khangvinh4656 there already is an euphoria plug in for the unreal engine. The engine is extreme powerfull and offers you a huge amount of options. A real balancer, AI generated behaviours which are full configurable and you can combine them each for a huge amount of ingame situations. All without neural networks, without internet connection or high end hardware. Its the most advanced animation/physicsengine you can get until now ... The only problem with it is, that its not available because since zynga bought naturalmotion in 2013-2014, they stopped the distribution of ghe engine to 3rd party.

[@MidnighterClub]

Really nice, cogent introduction to neural networks too. The funny running man is just a bonus.

[@richerlecool]

Nice nickname mate Hyped for the new record?

[@Maxx__________]

Had this same thought, decided to look up if this is already a thing... and... yea. It sure is. Super cool! I can see this becoming standard when more dev tools and pre-trained animations are available.

[@kasp7674]

This is so cool. I'm looking forward to start seeing it in games in a few years to a decade.

[@anymyo9882]

you dont need to wait...look at gta4, max payne 3 or red dead redemption 1... the future was back in 2006 and called euphoria engine from naturalmotion...

[@damienscott6561]

@@anymyo9882 true, Euphoria is still ahead of it's time after more than a decade, but the side effect was a giant cost and difficulty of implementation. With latest motion matching breakthroughs, the same results could be achieved with simplier solutions,

[@anymyo9882]

@@damienscott6561 i agree. To send a developer from matural motion for implementing the engine and a specific behavior to any game was the reason that this great technology didnt become the standard. The better way would be to do it the same way like EPIC did it with the distribution of the UE4.

[@rickybloss8537]

I'm an indie dev and I'm currently working on a project with similar tech.

[@moussesucree]

@@anymyo9882 Not the same thing at all, it wasn't use for player controlled character, but for objects and ragdoll.

[@marcoleone393]

Basically what Rockstar has been using since 2005.

[@AllExistence]

He came to talk about procedural animation, ended up explaining structure of neural network for 30 minutes.

[@Tinfoilpain]

wait this isn't Two Minute Papers...

[@paranoidcake]

what a time to be alive!

[@nafurachan9061]

La Forge is seriously full of treasures

[@jimmini8951]

Exanima already features motion synthesis for character animations that doesn't require specific behaviours to be baked with machine learning, works on the fly with polygon soup, supports non-human forms (including adding and removing body parts) and generally just looks great. It's also used for much more than just visuals, as it allows a huge amount of control in combat, which always plays out differently.

[@JediEdi]

Damn. It's already been almost a year since that game's last update. So much unused potential.

[@NinponixTutorials]

32:34 this is cool

[@anymyo9882]

Looks like euphoria/morpheme engine from naturalmotion... after 10+ years finaly another dev studio was able to recreate something similar... keep in mind that ragdoll combined with physics based animations and AI behavior is something gamers wont miss anymore sience gta4, red dead redemption and max payne 3 but still getting disappointed from game to game...

[@damienscott6561]

I don't know why, but for some reason at some point developers stopped focusing on making the environments in videogames interactive. I still remember having enormous amounts of fun while playing Max Payne 2 and Half-Life 2. Physics alone made these games insanely replayable and pleasant to interact with. Nowadays the environments are very beautiful, but they are so static that it doesn't feel like you are totally present in the situation.

[@AlexFord-gp7by]

@@damienscott6561 things can easily get out of control if every thing was procedurally generated like buildings destruction, animations, ai, etc that's why we don't make it procedural a yearly release game like assassin's creed would take 5-7 years to make if everything was procedural it's not worth it and will kill the performance you can see examples of procedural AI Getting out of control in games like Oblivion and sadly natural motion stopped selling Euphoria/morphine to game companies due to no one wants it or want to use it even Rockstar themselves Don't use it anymore so keeping everything under control will give way better results and much less time

[@khangvinh4656]

Euphoria was amazing only if you are able to pay millions to put it into your game and also having to develop the game around it making it extremely intensive for most hardware these days. The reason why most games don’t just include euphoria into their game is because that either they don’t have an engine that is supported fo it, too expensive, don’t have direct contact with NaturalMotion, and time consuming. This presentation showed a better method in dealing with AI animated ragdol that is more effective and easier to use that will hopefully become less intensive and implemented into games to come.

[@anymyo9882]

@@khangvinh4656 the expensive thing was not the licence itself, its the dev which was sent from maturalmotion to each game studio which was interested in using the engine and to create an euphoria config which fits to each project... there are some informations about it on the internet... its not really hardware intensive... it runs on xbox 360 and ps3 perfectly and its compatible to the most mainstream engines (unreal etc) related to the naturalmotion website... R* was just too lazy to configure it right for gta5... atm im working on a mod for RDR2 where i reconfigure the euphoria engine to bring back the cineastic look from RDR1... its a cool look behind the scenes and you have a extreme amount of options... what ive created so far is not comparable to RDR2 vanilla ragdoll... its complete different and unlocks much more reactions and behaviours but its a little bit time consuming...

[@khangvinh4656]

@@anymyo9882 you just explain why it is expensive. Having personnel sent to a company in working long time and paying for all of that is costly. And yes, the euphoria engine is very CPU intensive. And no, euphoria aren’t compatible with the unreal engine unless it is implemented. It has been a misconception even that Euphoria is just a feature that anyone can create or mod to a game but rather it has to already be in the game files for that specific game to have euphoria best example, gtaV. It is cool that you have created such a great and amazing program that can mimicked or behave like Euphoria so much faster than most developer and it would be great if you would join the team that is in-charged of this project as it would greatly help development.

[@endoflevelboss]

"so hello". So indeed.

[@reionami5983]

Just put this in the games already

[@anymyo9882]

you can find in gta4, red dead redemption 1 or max payne 3... actually its old stuff

[@acidrain9924]

Anym Yo Yeah old, but still very complex afik only NaturalMotion was able to pull it off in a convincing manner

[@grimsk]

virtual robotics 라는 게 있군요

[@filipex2000]

Amazing!

[@DTKenneth]

Anyone else think this was TABS at first glance of the thumbnail?

[@LongmireLocomotion]

Love

[@Sebaxterr]

♥

ayy longmire fancy seeing u here

[@soaringsquid0.129]

Every American student's relationship with their homework: 0:07

[@AntonQvarfordt]

2:02 Well... No.. He's just as not there as before.. But it's sure starting to come close to me thinking and feeling that he is :) He also has this thing where he looks like something in a farcical play, playing the aloof idiot :P I think it's a really good implementation of an actual person with some quite significant neurological damage affecting motor skill :P Like he actually have to make sure he did indeed take that step and plant his foot by really exaggerating everything step.

[@SirSpence99]

That was some good entry level information hidden by a terrible presentation with an atrocious label. It needed to be practiced a few more times and something like "Machine Learning Summit: Intro to machine learning with ragdolls" would have been a much more accurate title.

[@aigen-journey]

To be honest that was not a good introduction to RL. He confuses a lot of terminology there, for example that chart at 16:00 is just nonsensical. There are really great resources for entry level RL, like lectures from DeepMind's David Silver, or Stanford's CS234, Spinning Up from OpenAI ...

[@aigen-journey]

Later in the talk again ... credit assignment is a different thing than discount factor and he describes this like it's the solution to CAP ... I kinda see the confusion here, but really this is not well explained

[@clavetsi]

@@aigen-journey Sorry you didn't like my talk. I had to take dramatic shortcuts to reach the policy gradient equation in 20 minutes, for an audience of non-specialists. I knew some people would not agree with these shortcuts. And by the way I stole the chart from an authority in modern RL: kzbin.info/www/bejne/iZDKqIygZrZgapY

[@homanchou]

I thought the talk was informative and inspiring. I have a question though. How did you search for a motion that matched the current state? Was that with some kind of vector database? I’m also curious if you needed to normalize the current pose into some local coordinate space.

[@anymyo9882]

Naturalmotions euphoria engine can do it without neural networks 😂

[@AnimationESP]

It's not the same, this is about finding solutions for locomotion using Motion Matching. Not about applying physics to a ragdoll. Euphoria is amazing don't get me wrong, but this is not attempting to achieve that level of physics reactions, but to get a physic ragdoll move using motion capture data only on a changing environment.

[@ViceZone]

Euphoria indeed uses AI

[@anymyo9882]

@@ViceZone Thats a matter of definition. Its no AI like people think of teaching a RIG to balance. Its to 100% configurable by file. There is nothing like self learning or something like that. So its not using AI.

[@ViceZone]

@@anymyo9882 There's many types of AI, I don't know exactly how it was made, but with a little google search you can find that euphoria uses some kind of custom made AI.

[@anymyo9882]

@@ViceZone its scanning the environment and links the corresponding task to it. Thats it. AI.

[@WhenTheSunLordsFell]

God he’s bad a public speaking

[@DTKenneth]

Anyone else think this was TABS at first glance of the thumbnail?

[@Bearbones]

No