This post kicks off a series of retrospective posts from game|tech, a small but focused 2-day seminar held in 2004 about Creating Believable Characters. The event was organized by industry veterans Jonathan Blow and Jeff Lander with the help of Chris Hecker. Many of the sessions are relevant for those looking at the low-level aspect of AI & animation integration.
The first talk chosen for this post is entitled “Dynamic Motion Control” by Torsten Reil of NaturalMotion, and contains a fascinating look at the trials and tribulations of creating animation based on a genetic algorithm to evolve neural networks. In particular, modeling biologically-inspired actuators which control movement (in the talk, walking) proved difficult but ultimately successful enough to commercialize.
Figure 1: Leg actuators controlling a walk.
The talk mainly emphasizes how difficult the area of physics-based motion synthesis is, and that starting it from scratch is very hard to accomplish. NaturalMotion has made steady progress since this presentation and is now able to get its animation technology into certain videogames where physics interaction plays an important role.
The audio is below, which is good to listen to along with the slides, although the videos are missing.
Dynamic Motion Control Torsten Reil Download MP3
Dynamic Motion Control Torsten Reil, NaturalMotion Download PPT
NaturalMotion’s euphoria Technology, editorial on AiGameDev.com
If you have an comments or thoughts about this kind of technology, and wish to discuss the ideas in the talk, feel free to post them below!