Found a video showing AI weaving animations. Applicable to gun and blade?

[Uthael]

 

[MillbrookWest]

These are largely academic (which is where the channel gets its name). Taking this and implementing it becomes it's own thing unto itself. It's cool that we're heading that direction at least, but not something you'd hold your breath for in the short term.

[Rawbeard]

ok, but what would be get out of this, other than "we have cool AI using up a lot of resources"? 

[MillbrookWest]

45 minutes ago, Rawbeard said:

ok, but what would be get out of this, other than "we have cool AI using up a lot of resources"? 

Animation in video games is usually "canned". That is: an animator has either defined strictly the animation set that can be used, or defined loosely the set on animations that can be interpolated/blended between. In either case, it's a lot of work, and this ignores things like IK.

The theory here is that an AI drives the animation entirely, lessening the workload as the animator does not need to generate an anim to cover every conceivable state a player might find themselves in.

Once the AI is trained, you could show it a whole new level it has never seen before, and it will deal with it in real time. Whereas now, if you generate a whole new level an animation team has never seen before, the team has to get to work creating animations that work in the environment (this of course implies the two levels are very different from each other... walking on a flat plane in one is walking on a flat plane in the other).

A better example could be as small as a player stepping on/over a box. Excluding IK, if an animator defines an animation where a player steps up on a box, for every instance of a box the player can step up on, the same animation plays. Meaning every box you can possibly step on has to be of the same size, or else the animation no longer aligns with the size of the box (if it's too small, the player steps up onto thin air etc.). If the AI is trained to "step up on boxes", you could then have any box and the AI would drive the animation to match the size of the box.

Edited October 10, 2021 by MillbrookWest

[Rawbeard]

ok, great, but what is the cost of having AI run animation? will the game even run on a console? what type of hardware does a PC need to make it work?

[MillbrookWest]

7 hours ago, Rawbeard said:

ok, great, but what is the cost of having AI run animation? will the game even run on a console? what type of hardware does a PC need to make it work?

At one point in time, many of the render techniques currently in use were also part of the "will it even run" philosophy. Hardware is always getting better, and optimizations are made to techniques to decrease their time complexity. SSAO was once the domain of the highest of the high end PC systems, yet the PS360 ran them fine all the same, eventually.

As i said above tho, you wouldn't hold your breath for this. There are far larger issues to solve with implementation than simply its cost to run.

[Uthael]

Performance-wise, if it would be applicable to a sports game and is able to run in real time, reacting to several button presses, it's probably quite optimized.

In my opinion, novel ideas take priority over fixing minor bugs and game balance. That's why Warframe IS Warframe in the first place.