Making animations to use in video games, especially in 3-D, is a really long, difficult task. I tried to do some animation work a little while back, and I found this the hard way. Knowing how to make walk, jump, and many other animated cycles takes a while, and getting them to interact with the environment correctly is really hard. This is why researchers at the University of Edinburgh are looking to use neural networking, along with motion tracking, to make more realistic, smooth movements.
Motion capture has been used in games and animated films for a while, using real-life footage as a basis for the movements of a character, but it can only go so far. You need animations for every possible combination of moves, such as, as the article says, climbing up stairs while crouching and drawing a bow. In the current method of animating, animations would have to be made for each unique scenario, but neural networks can fix this. The program adjusts the movements based on the input given, and the data based system makes the animations fit.
Even though this system is very complex, and likely won't be implemented in anything more than a tech demo for a while, it shows that this technology can be used in animation, and that should make all animators breath a sigh of relief for the future.
Original Article: http://theorangeduck.com/page/phase-functioned-neural-networks-character-control
No comments:
Post a Comment