571 views

4:51

Watch
- Search Videos
  
  Advanced search »
- Submit a New Video
Discuss
About
- Try clicking on the various channels to see videos posted to that channel - if you're into music you can click the Music channel to get more music sifts - go on and explore! Use the "hotness" when you have opened the channel link to get videos that have gotten a lot of votes, or use "newness" to have a look at the new sifts.
  When browsing sifts you like, have a look at the Related Posts list to see other related videos, click the tags under the video title to see more sifts tagged with the same tags or have a look at other sifts from the same user in the More Posts By list. If the sifted video is in a play list, have a look at the play list to see other interesting videos the Sift users have linked together.
  Enter the Sift Lounge to chat with online sifters about anything and everything, or read interesting posts in the Sift Talk forum.
  The Top New Videos list is always full of interesting, current videos. It is always recommended that you have a look through the Top New Videos, but also have a look at these:
  - Upcoming Unsifted Videos - active video submissions that neither are published to the front page nor archived to the submitter's personal queue
- Have you seen a really funny, surprising, scary, interesting, cute or in any other way extra good quality video?
  Post and share it here - VideoSift is the best video aggregator on the Web, where we share, discuss, promote, and interact with the best videos out there on the web. Post it here, and share it with the large community of video-interested users. And - you are guaranteed to find more very good videos here already!
  Mind you, we only share videos you discover that someone else has made; if you "self link," by posting a video you have made, have posted or have interests in, you risk the ban hammer - read the posting guidelines before posting. The sift is NOT for promoting your own videos - it is for sharing videos others have made.
  Submit a Video Now »
Login

Neural Networks for Character Control

submitted by Ashenkase

7 years 7 months ago

A real-time character control mechanism using a novel neural network architecture called a Phase-Functioned Neural Network.

The entire network is trained in an end-to-end fashion on a large dataset composed of locomotion such as walking, running, jumping, and climbing movements fitted into virtual environments.

The system can therefore automatically produce motions where the character adapts to different geometric environments such as walking and running over rough terrain, climbing over large rocks, jumping over obstacles, and crouching under low ceilings.

4 Comments

Payback

says...

You know you have a *quality algorithm when watching the output makes you sympathetically tired.

siftbot

says...

Boosting this quality contribution up in the Hot Listing - declared quality by Payback.

artician

says...

Really impressive, but despite the description I'm not sure how this differs from the current, modern approach, beyond using more processing power for the environment assessment and realtime decision-making, weighting and blending. It just seems like a more thorough and complete AI-supported locomotion system.