Thursday, May 05, 2005

Realism of computer games

As of 2005 we are on the threshold of realism in games. It is finally possible to simulate certain aspects of reality in real time and with sufficient precision to declare it an accurate simulation overall.

For example, the Forza Motorsport racing simulation for Xbox is physically realistic. It is mostly on par with reality, even though it's not indistinguishable yet. To achieve this, programmers from Microsoft Game Studios take into account between 3000 and 10000 variables and simulate all aspects of driving, running the simulation at 240 ticks per second. For Race Against Reality Popular Science asked a veteran gamer and a professional race driver to extensively test drive both real cars and their virtual prototypes. The conclusion was that the game simulation is accurate.

Similar level of realism is available for the flight simulators, again from Microsoft. Some simulators are so realistic that pilots are allowed to log the virtual hours just like the real ones.

However, these simulations are not completely realistic yet. There are still things that can be improved though before we have perfect VR.

  • Graphics aren't perfect yet. One of the bigger problems is lighting and shadowing. To make realistic materials technologies such as RealReflect need to be developed.
  • Sound - there is still no good programmatic sound generation. It's all samples, mostly.
  • Global physics - it's possible to simulate several objects (cars, planes) very accurately, but an all-encompassing simulation is still too complex for the tech we have.
  • Simulation of acceleration, tactile contact and everything else related to physically "being there".
  • AI to make the world come alive

The video-realistic graphics based on general-purpose stable rendering systems (i.e. no more custom-made rendering engines for every new project) will come around 2010-2015. The programmatic sound may be delivered somewhere between 2015 and 2025. Global physics may be done sufficiently well around 2015-2020. The realistic simulations of all senses may come somewhere between 2015 and 2025. Sufficiently good non-human and domain specific human AI (i.e. for an NPC that can realistically perform in a narrowly defined context) may come somewhere about 2015-2020. Good human-level AI (in the context of video games it's a companion that you can interact closely for many hours in a variety of situation, including free-form talking) is a more complex problem and will probably not be achieved until the 2030s.

Still, we have already entered the realm of virtual reality. In some aspects, although not in all, virtual environments are already as good as real ones.

No comments: