[x3d-public] Story-to-Motion
GPU Group
gpugroup at gmail.com
Wed Nov 22 19:36:45 PST 2023
Looks interesting -- go for it.
And maybe some of the papers referenced might have some interesting
algorithms.
Relating to HAnim, could there be a HyperMotion node, and .motions [Motion]
would be the 'database of short animations' the article mentions, and with
any attributes needed added to the Motion node, for example 'name' for a
name lookup / matching of motion, a 'cycle duration' handy, distance per
cycle etc.
HyperMotion might compute or be programmed with a queue of (name,
location_target(x,y), duration(time)) tuples, and when it gets to a target,
it looks up the next motion 'name' from the .motions list, and does a
transition blending for a fraction of a second, between old and new motion.
But the story-to-motion or whatever else is generating the queue of tuples
- that's the part that looks either laborious pre-programmed, or requires
AI automation of some sort to generate the queue.
And what about collision avoidance with other animated characters and
static scene elements? What about switching to stair climbing animation
when over stairs and other scene-responsiveness?
Reminds me a bit of the Crowd Particles function map.
https://freewrl.sourceforge.io/tests/40_Particle_systems/crowd/function_map.png
https://drive.google.com/file/d/1krC0k2CKZDuOGgzlyaFosrFbjFRO2qzx/view?usp=drive_link
which could be done other ways such as function bounding boxes / rectangles
or as the paper suggested trajectory lines.
So HyperMotion might have children field with some sort of (zone, function)
tuple list it can use to switch between motions along a trajectory.
-Doug
On Wed, Nov 22, 2023 at 7:36 PM John Carlson <yottzumm at gmail.com> wrote:
> Hot air:
>
> What may be more interesting to game developers is tracking many, many
> players in a game, and then producing NPCs based on traces capable of
> playing like humans. It’s similar to “predict the next word” for popular
> AIs. Then add reinforcement learning to improve successful NPCs. This is
> where synthetic motion generation (ala Story-to-Motion) may come in handy
> instead of relying solely on humans for training data. The question is,
> can story-to-motion provide enough creativity?
>
> On Wed, Nov 22, 2023 at 7:57 PM John Carlson <yottzumm at gmail.com> wrote:
>
>> Please let us know of supporting authoring systems and players.
>>
>> From what I can see of the video and paper, the Story-To-Motion is more
>> about position of the humanoid and rotation of joints. If these are
>> “Group”ed by keyword, and smooth transitions are accomplished by chaining
>> animations, then some degree of automation can result. If result numbers
>> are available in a potential tool like JAminate, Blender or a text editor,
>> and accessible to human animators for optimization, all the better!
>>
>> I’m thinking more and more about parsing interpolators into JAminate,
>> which has been a goal. Using JAminate or a spreadsheet, one could
>> potentially group key-fractions and keyValues, then output interpolators
>> and routes could be published in addition to chaining of moves with Boolean
>> Sequencers.
>>
>> John
>>
>> On Wed, Nov 22, 2023 at 7:18 PM Joe D Williams <joedwil at earthlink.net>
>> wrote:
>>
>>> It might be that the closest x3d going on with this is in web3d x3d
>>> HAnim WG facial animation standards-track. The facial mesh of various
>>> densities are organized into regions activated in certain sets to perform a
>>> certain expression. So, standard keywords and parameter sets can activate
>>> standard related areas of the mesh to allow use of standardized animations
>>> as well as serve as a prime basis for the personalization of the Humanoid
>>> that you really want to do. This is not a quick and simple effort. It is
>>> very interesting, design-friendly, much simplified authoring, and much more
>>> standardizable and demonstratable at all levels of expression using the
>>> hanim displacer concept.
>>>
>>>
>>>
>>> Joe
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> -----Original Message-----
>>> From: John Carlson <yottzumm at gmail.com>
>>> Sent: Nov 21, 2023 3:19 PM
>>> To: Joe D Williams <joedwil at earthlink.net>, X3D Graphics public mailing
>>> list <x3d-public at web3d.org>
>>> Subject: Re: Story-to-Motion
>>>
>>>
>>> Here’s a better link i hope.
>>>
>>>
>>> https://arxiv.org/abs/2311.07446#:~:text=A%20new%20and%20challenging%20task,level%20control%20(motion%20semantics)
>>> .
>>>
>>> This should be a better link (added f):
>>>
>>> https://arxiv.org/pdf/2311.07446.pdf
>>>
>>> Apologies for multiple posts.
>>>
>>> John
>>>
>>> On Tue, Nov 21, 2023 at 5:08 PM John Carlson <yottzumm at gmail.com> wrote:
>>>
>>>> https://arxiv.org/pdf/2311.07446.pd
>>>>
>>>> Has this research been peer-reviewed?
>>>>
>>>> It’s hard to read on my cell phone, I may get my computer and attempt
>>>> to read it.
>>>>
>>>> John
>>>>
>>>
>>>
>> _______________________________________________
> x3d-public mailing list
> x3d-public at web3d.org
> http://web3d.org/mailman/listinfo/x3d-public_web3d.org
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://web3d.org/pipermail/x3d-public_web3d.org/attachments/20231122/db84e62d/attachment.html>
More information about the x3d-public
mailing list