Playable Tech Demo Shows Why Future Games Should Be Animated Using Neural Networks

Image: Sebastian Starke

Visual fidelity in games has improved massively in the last few years, to the point where ray-tracing is now being used to complement rasterisation in real time. Animation, however, has lagged behind, but a combo of motion capture and neural networks could change that.

If you ever want a sneak peek at the technology games will be using over the next decade, just check out SIGGRAPH. For instance, at this year's conference, University of Edinburgh animation and AI student Sebastian Starke will be showing off this neat bit of work.

Called "AI4Animation", the system "can produce natural animations from real motion data" via "a novel neural network architecture, called Mode-Adaptive Neural Networks":

Instead of optimising a fixed group of weights, the system learns to dynamically blend a group of weights into a further neural network, based on the current state of the character.

That said, the system does not require labels for the phase or locomotion gaits, but can learn from unstructured motion capture data in an end-to-end fashion.

The final product is smooth, realistic animations, minus the painstaking work of doing everything by hand. Interactive demos are available for Windows, macOS and Linux, so you can play around with it yourself.

Image: Supplied

In other news, I think we're about ready for a new Okami, Capcom.

AI4Animation [GitHub]


    Now apply neural networks to how NPC's interact with the player, not just in combat but in dialog and general livable world sense. :)

    Last edited 01/07/18 8:00 am

    Some of that hind leg work is real nice. I'd love to see this crossed with something like Euphoria.

    That's real pretty and all, but they have totally messed up the way in which a dog places weight with it's hind legs and toes. Dogs have a completely straight ankle all the way through to the tips of the toes. The only time a dogs has a "foot", so to speak, is when the animal puts weight on it. The hound in that demo looked really janky and showed that, despite observing canines in motion, they still can't replicate actual quadruped motion, real-time, accurately.

Join the discussion!

Trending Stories Right Now