Games By Matthew Humphries Jan. 15, 2014 9:30 am
Capcom has made it clear that developing games for next-gen machines like the PS4 and Xbox One involves a lot more work, to the point where the company is developing an entirely new engine and toolset to speed up the whole development process. But faster development for more realistic looking and acting games can also happen through research and development within the different specialities that exist in games creation.
One work-heavy area is character animation and the associated motion capture many games rely on. Humans can be motion captured to get animation data for hundreds or thousands of actions relatively quickly, but that still limits how realistic their reactions are to unpredictable in-game events. If a game uses characters that don’t actually exist, or it’s very difficult to motion capture them (e.g. animals), then the animation is done by an animator and by hand. You get great results, but it takes a long time.
Research being carried out at Utrecht University is attempting to simplify character animation to the point where it almost becomes automatic and “for free.” Thomas Geijtenbeek’s PhD thesis entitled Animating Virtual Characters using Physics-Based Simulation replaces hand-crafted or motion captured animation with a learning simulation. The results, which have been implemented as an actual piece of simulation software with the help of Michiel van de Panne and Frank van der Stappen, are very impressive as the video below demonstrates.
What you are seeing is a range of different bipeds complete with muscle systems that learn to walk themselves over many generations of experimentation. The end result is not only a biped character that can walk unaided, but one that can react in an infinite numbers of ways as well as changing speed and coping with any type of walking surface. This is a far better solution than the static reaction animations most games rely on today.
Of course, such motion relies on a robust physics engine being in place and enough hardware performance to cope with not only rendering the game, but handling tens or hundreds of simulated characters moving around and reacting naturally. Even if that isn’t available today, the fact this research is being carried out and simulations already exist prove it works. In the not too distant future this could become the norm in video games and anywhere virtual characters are used.
There’s another, extended video of the bipeds you see above learning to walk which shows in much more detail how they learn and ultimately fail a lot. It’s an interesting watch, and also one that will make you laugh a few times. Watching different bipeds attempt to walk and failing is highly amusing.
More...
One work-heavy area is character animation and the associated motion capture many games rely on. Humans can be motion captured to get animation data for hundreds or thousands of actions relatively quickly, but that still limits how realistic their reactions are to unpredictable in-game events. If a game uses characters that don’t actually exist, or it’s very difficult to motion capture them (e.g. animals), then the animation is done by an animator and by hand. You get great results, but it takes a long time.
Research being carried out at Utrecht University is attempting to simplify character animation to the point where it almost becomes automatic and “for free.” Thomas Geijtenbeek’s PhD thesis entitled Animating Virtual Characters using Physics-Based Simulation replaces hand-crafted or motion captured animation with a learning simulation. The results, which have been implemented as an actual piece of simulation software with the help of Michiel van de Panne and Frank van der Stappen, are very impressive as the video below demonstrates.
What you are seeing is a range of different bipeds complete with muscle systems that learn to walk themselves over many generations of experimentation. The end result is not only a biped character that can walk unaided, but one that can react in an infinite numbers of ways as well as changing speed and coping with any type of walking surface. This is a far better solution than the static reaction animations most games rely on today.
Of course, such motion relies on a robust physics engine being in place and enough hardware performance to cope with not only rendering the game, but handling tens or hundreds of simulated characters moving around and reacting naturally. Even if that isn’t available today, the fact this research is being carried out and simulations already exist prove it works. In the not too distant future this could become the norm in video games and anywhere virtual characters are used.
There’s another, extended video of the bipeds you see above learning to walk which shows in much more detail how they learn and ultimately fail a lot. It’s an interesting watch, and also one that will make you laugh a few times. Watching different bipeds attempt to walk and failing is highly amusing.
More...
