• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Walking simulation just got a whole lot more interesting

a8H3f2e.gif


FMmAjSu.gif


PA0cfFP.gif

This video is fantastic.
 
Eh, I'm going to be completely honest; I only came in here because I read the title as 'Wanking simulation'.

Sorry for wasting your time, but screw you OP! You got my hopes up.

(It's funny now that I think about it. Why would I need a simulation for something I can do any time I please and something I did not two hours ago? Now I feel stupid.)
 
Dat Gen 80...

So funny, love what they are doing here, would like to see how this is possible with gameplay and what kind of impact it would have on processor ...

Also, box to face = dead
 
TucoBenedictoPacifico said:
I... don't think you understand how evolutionary algorithms work, in principle.

They don't go through different generations "on the fly", in real time.
You build a large amount of entities, usually almost randomly at first, then you select the very few that work SLIGHTLY better toward a goal you decided, let them replicate making small changes and you test them again in generation #2.
Then you select once again those few which work better, you change them only slightly, and you test again in gen #3, and so on.
By the time you reach gen #999 (or whatever number you are fine with) you have an entity which has eventually evolved in something surprisingly efficient toward the goal you set (fighting, moving quickly toward a direction, grabbing an object and moving it around, etc), and without needing any input from you.
The "scary" thing about it is that in the end you have something that works and you don't even necessarily understand how it does what it does.

But then you save just the final product and use it in the final software, you don't have to go through all those 999 generations each time you need to use that entity once again.
And *surely* not in real time.
Click to expand...

The only thing worth adding to this is that, dependent on what has actually been done, the 'evolved' controller may or may not be designed for a very particular scenario (and hence not be particularly useful for general scenarios).

I remember reading a paper from a few years back in the area that suggested to me the controller created would be useful only for the task it was trained on. For example, if the task was 'kick ball into goal' it would become very good at that, but change the position of the ball or goal and it would be completely useless. That case was an extremely simple simulation which learned a periodic function for each joint.

I'm not sure on the simulation in the video, but there is a huge difference between a controller that acts based on sensors (generalised) to a controller that learns how to deal with a specific scenario. If it is the former, then this is extremely exciting indeed.
 
This is so awesome.

If I understand things correctly, it is the generation of efficient moving patterns that is computationally intensive, not running the final simulation. The tech should have obvious uses in open world games to give 300 or so bystanders pre-baked but unique walking patterns based on their weight/proportions. Could also be interesting in 3d exploration games with automatic generation of diverse animal like enemies (like a 3d version of the system in Starbound). Aside from games, the tech should be very useful in advanced custom made prosthetics.

Ubisoft and Rockstar should licence the tech asap.
 
TIGER-COOL said:
I just watched that entire video. 5 and a half minutes of walking simulations. I sat there and watched it. You know what the worst part is? I loved it. Every second of it. Especially the guy on the bridge getting hit with cubes. I liked that part the most.


Well, goodnight.
Click to expand...

Don't fell bad. I'm doing the same thing. 5:34 a.m.

I regret nothing.
 
I really wish they'd make a game about this. BoxCar kind of does this, but a mid-tier level game about evolution and optimization by the computer could be fantastic and addicting.
 
PA0cfFP.gif


Yup, this is ready to go for console wars. Insert console names and your favourite faces.

Good work here. The smaller blocks looked like they were pretty light. Would be more interesting to see them with adding some weight on each block to see some progressive adaptation, adjusting stance to prepare for landing (or impact if they're feeling fancy), and so on.

Takeshi's Castle AI challenge with randomly generated levels would be great fun to watch I think.
 
sn00zer said:
https://vimeo.com/79098420

From what I understand it a simulated muscle system that learns how to walk based n muscle configuration....basically you throw a rig at it with a given weight distribution and it will figure out how to walk/ run ...really interesting results ....no explanation as to computation time so this may all be years off
Click to expand...

I love this stuff!

I would think that the "generations", might be for refined muscle memory, tuning. I wonder if that can be pre-baked and then just loaded up for each specific model config, so runtime calculations are reduced.
 
This looks to me to be software to streamline animation. It's not meant to run in real time during a game. Now animators dont have to build animations by hand for most things once the software has the proper animations for the skeletal and musculature design figured out. And since it can work for any skeletal design it can build the animation foundations for humans and animals alike. It would free up time for animators to concentrate on more custom animations.


This stuff isnt new though. This is what the very first videos for natural motion were showing off. Only I think their system started off with dozens of models attempting to balance and walk correctly and then the animator would remove the worst performing half and then run the simulation again over and over until the software zeroed in on the correct balance and movement for the skeletal structure.
 
Euphoria have done similar research on this subject. While there might be some niche area where this is applicable for games I think artists mostly prefer to animate things by themselves to get the look that they are aiming for (unless that would be impractical or impossible).

The really interesting part is the area of application in advanced robotics.
 
a.wd said:
Dat Gen 80...

So funny, love what they are doing here, would like to see how this is possible with gameplay and what kind of impact it would have on processor ...

Also, box to face = dead
Click to expand...

They should have stopped at Gen 80, clearly perfection had been reached, so any further improvement is a sham.


PA0cfFP.gif


That's just the goofiest funny ass stride I've ever seen, and the way he veers off makes it even better. I need a Gen 80 walking simulator game where I can lay out obstacle courses for that guy.
 
FloatingDivider said:
PA0cfFP.gif


That's just the goofiest funny ass stride I've ever seen, and the way he veers off makes it even better. I need a Gen 80 walking simulator game where I can lay out obstacle courses for that guy.
Click to expand...
Every time I see this GIF I laugh all over again. That Gen 80 creature is amazing! So funny how he desparately tries to follow the Gen 999, but with all the swaying neck action the poor guy eventually just veers off.
 
TucoBenedictoPacifico said:
That's actually the same mix of physics simulation and evolutionary algorithms used in the Euphoria Engine (GTA, RDR, Max Payne 3, etc).
So yeah, it can have a lot of applications in games. It's just that usually is considered something too expensive and complex for most studios.
Click to expand...
It would be really interesting to see generative AIs inside of an engine like Unity.
 
TucoBenedictoPacifico said:
I... don't think you understand how evolutionary algorithms work, in principle.

They don't go through different generations "on the fly", in real time.
You build a large amount of entities, usually almost randomly at first, then you select the very few that work SLIGHTLY better toward a goal you decided, let them replicate making small changes and you test them again in generation #2.
Then you select once again those few which work better, you change them only slightly, and you test again in gen #3, and so on.
By the time you reach gen #999 (or whatever number you are fine with) you have an entity which has eventually evolved in something surprisingly efficient toward the goal you set (fighting, moving quickly toward a direction, grabbing an object and moving it around, etc), and without needing any input from you.
The "scary" thing about it is that in the end you have something that works and you don't even necessarily understand how it does what it does.

But then you save just the final product and use it in the final software, you don't have to go through all those 999 generations each time you need to use that entity once again.
And *surely* not in real time.
Click to expand...

Also, the resulting algorithm / behaviour does not have to be computationally intensive for what it does. The computationally intensive part comes from having to test those several entities every generation. Also, the, being computationally "light" can also be selected as a fitness criteria.
 
hesido said:
Also, the resulting algorithm / behaviour does not have to be computationally intensive for what it does. The computationally intensive part comes from having to test those several entities every generation. Also, the, being computationally "light" can also be selected as a fitness criteria.
Click to expand...
Yeah, there's that too.
And I guess you can eventually decide which one, among different selective principles, must be prioritized or put in background (i.e. "walking nicely is more important than being computationally light").
 
About as interesting as Rockstars' Rage engine and Euphoria's natural motion.
Which is king! GTA4, 5 and not to mention Max Payne 3 are never the same thing over and over again thanks to this.
 
You must log in or register to reply here.

Similar threads

Quest 64 just got a recomp for PC; your day just got better! or worse?
21
493
TintoConCasera replied
Id Software on celebrating its 35th anniversary: "We've got a lot coming up"
Event Game Dev  2
80
1K
MiguelItUp replied
NieR: Automata creator Yoko Taro says he’s actually been working on a lot of projects recently, it’s just that they got cancelled midway
4
417
Signalis replied
Young Americans Are Spending A Whole Lot Less On Video Games This Year
2
83
3K
MALpractice311 replied
The nearly 20-year-old PS3 just got another system update
News Retro Platform  2
66
910
Krathoon replied
Top Bottom