Motion Matching - 'Dance Card' Breakdown

[@EricJacobusOfficial]

Fascinating seeing how you put all footwork iterations into one dance card. Thank you.

[@mazew88]

Thanks so much for sharing this. Since seeing your GDC talk, I've been wondering what was involved in the dance cards. I've been finding your posts on this channel really educational, and looking forward to seeing more.

[@KristjanZadziuk]

Thanks Daniel, feel free to share this channel i plan on keeping up with the video game insights :)

[@weltraumimport]

Amazing concept

[@zeyadjy7229]

Thanks so much for sharing

[@KristjanZadziuk]

themarkos zizo kzbin.info/www/bejne/gYS3n2abhamna5I

[@sahilpohare]

Can you provide dance card mocaps for learning purposes.

[@LawrenceHarding]

I absolutely adore the work you and your team have done here. It truly is revolutionary, so thank you all for the advancement in animation. I do have a few questions, if you get around to seeing this. 1. How large is each individual dance card, file size wise? Raw Capture seems heavy, so I'm curious how you were able to incorporate so much animation into For Honor, with such a small allocation of RAM, given the animation budget on an XB1 and PS4 game is usually like 200-250MB or so. 2. Also, when you do a dance card specifically for just say, walking, do you have to have an entirely separate dance card of walking with a sword or a gun? I don't understand how the system would be able to decipher between "normal walking", compared to walking with a sword or gun by your side. It seems like in the node graph, you'd have to switch between those datasets when holstering and unholstering your weapon. So the way I'm looking at it, you're essentially re-doing 90% of the "normal walking" dataset, simply to now walk with your weapon out. That seems like it could become a problem, especially if the file sizes are large. Either way, thank you again for the work you have done. This is easily going to power the next generation and hopefully allow smaller devs to create incredible animation, thus helping bring down these ballooning dev costs.

[@jonasmlgaard8884]

Where can i watch the GDC talk you did? Super excited to try build a prototype using this approach

[@KristjanZadziuk]

+Jonas Mølgaard it's available on the GDC vault (links in the description) but its behind a paywall currently, the pdf is free though. I'll look to release it for free later on in the year

[@jonasmlgaard8884]

Ah i see.. I dont think i wanna wait to hear all that goodness :-) getting a gdc subscription is probably a good investment anyway so i can keep up with what all you pros do :-D again super excited about this technique, and seems totally different than the spaghetti statemachine stuff i've done, but also doable once the selection parameters and datastructures are in place

[@KristjanZadziuk]

Good Luck! I will be posting more video game related insights, intending on building this channel much more.

[@jonasmlgaard8884]

Thanks. That sounds awsome man. I subscribed and will keep an eye out

[@KristjanZadziuk]

Thanks man, appreciate it

[@MementoMori000]

Great work! Really inspiring. A couple of questions: How much (if at all) was this technique inspired by this work? grail.cs.washington.edu/projects/motion-fields/ Would you agree that the real key in the success of your work here was in the creation/iteration on the dance cards?

[@KristjanZadziuk]

Memento's Lair thanks, to answer you questions 1. Yes this is inspired by that paper, our programmers have made no secret of that, it was adapted to be of use to video games. It needed more work to aid response times. 2. The dance cards are part of the planning and give more structure to the technique and allow animation teams to understand what they are going to get. It is a big part of the solution but not the only part. With out the dance cards its just a mess of data. This was an effective way to capture everything needed. Hope that helps

[@MementoMori000]

Thanks!

[@mehdisaffar]

Thank you so much for sharing. I have a question: can this system be adapted for more sophisticated movement? Like say a climbing system?

[@swinny_]

not claiming to be an expert, but with enough motions + ik for contact points I dont see why not