New Approaches to Physically Based Computing
Author
With today's demand for ever increasing realistic physics in games which then leads to increased CPU compute time, how are game developers using new ideas to produce "better" physics with reduced compute times? Any Tips or new upcoming approaches beyond sphere - sphere collision detection? Sample rates? Fuzzy Logic? Anything? I would like to attempt to implement one new upcoming approach to increase physics while reduced CPU time.
Thanks in Advance.
Most games do not use most of good old physics simulation tricks(i mean speed-up tricks) just only because "hard to implement", so....
Quote:Original post by Dmytry
Most games do not use most of good old physics simulation tricks(i mean speed-up tricks) just only because "hard to implement", so....
So the issue boils down to increasing the budget again, hehe [smile]
Some kind of physics level-of-detail would (might) be interesting - i.e. increasing the simulation accuracy where the user can see, and decreasing it elsewhere (without interesting artifacts).
Designing an API so that you can (internally) split the simulation up into completely separate "islands" that can be farmed out to multiple processors (bearing in mind the next generation of consoles).
There's an amazing video here:
http://chihara.naist.jp/people/2003/takasi-a/research/
showing 3D fluid dynamics mostly running on the GPU. Not sure how useful this would really be in a game though, since the GPU is probably needed for other things!
Designing an API so that you can (internally) split the simulation up into completely separate "islands" that can be farmed out to multiple processors (bearing in mind the next generation of consoles).
There's an amazing video here:
http://chihara.naist.jp/people/2003/takasi-a/research/
showing 3D fluid dynamics mostly running on the GPU. Not sure how useful this would really be in a game though, since the GPU is probably needed for other things!
the GPU can be used for many things, even audio processing. Now with the PCI-Express, it will be something to look out for.
Physics level of detail kind of exists already. It's not uncomon to use some basic physics for bots and other simpler things, or things far away from the viewier.
there are a couple of interesting ideas kicking around, which are a bit out of the ordinary. Verlet particle systems being one, but they don't work that well really. THere is the continous collision detection approach, which is costly (see Stephane Redon stuff). Most of the physics engines now go as far as rigid bodies. There isn't a full-on soft body system that I know of and that works.
There is the finite elements approach, more like mechanical engineering. very costly. All the fluid dynamics is barely touched upon, which would be cool for things like fire, wind, clouds formations, water, ect....
Inverse kinematics, which already exists in some games, vastly improves the animation and the suspension of belief in games (seing characters walking in midair, with arms sticking out of walls isn't exactly accurate), but it's obviously not that easy.
breakable scenery is also not widely used, because most of the time it brakes the gameplay, and is quite costly/difficult.
there is also the research on A.I. assisted animation and physics, which apparently gives very impressive results.
All these systems are complex, and there are not formalised, which tends to put brakes on the development.
Hardware-accelerating physics is not as easy as rendering, or the current state of the rendering we use, which are at the monent just polygon pushers. That can be standardised nicely, and stuff streamed into big fat pipes, whereas physics in general is a lot more open ended, and does not fit a pipelined architecture so well.
I think given the technology, we can do much more than what we currently have. Professional physics engines are already pretty optimised, they just lack some features that could be implemented with a lot of time and effort. Now that dedicated physics engines are a reality, I think we'll see more and more features appearing in the games. Already, the features in UT2004, Half Life 2 and Flat Out (full on rigid body dynamics, with hundreds of objects interacting, breakable objects, ragdolls) are pretty common in games, and will be the basic stuff for the next generation. But even that requires a LOT of know-how and work. Maybe in a couple of centuries, they'll teach rigid body dynamics, constraints, lagrangian dynamics to 4th graders :)
Physics level of detail kind of exists already. It's not uncomon to use some basic physics for bots and other simpler things, or things far away from the viewier.
there are a couple of interesting ideas kicking around, which are a bit out of the ordinary. Verlet particle systems being one, but they don't work that well really. THere is the continous collision detection approach, which is costly (see Stephane Redon stuff). Most of the physics engines now go as far as rigid bodies. There isn't a full-on soft body system that I know of and that works.
There is the finite elements approach, more like mechanical engineering. very costly. All the fluid dynamics is barely touched upon, which would be cool for things like fire, wind, clouds formations, water, ect....
Inverse kinematics, which already exists in some games, vastly improves the animation and the suspension of belief in games (seing characters walking in midair, with arms sticking out of walls isn't exactly accurate), but it's obviously not that easy.
breakable scenery is also not widely used, because most of the time it brakes the gameplay, and is quite costly/difficult.
there is also the research on A.I. assisted animation and physics, which apparently gives very impressive results.
All these systems are complex, and there are not formalised, which tends to put brakes on the development.
Hardware-accelerating physics is not as easy as rendering, or the current state of the rendering we use, which are at the monent just polygon pushers. That can be standardised nicely, and stuff streamed into big fat pipes, whereas physics in general is a lot more open ended, and does not fit a pipelined architecture so well.
I think given the technology, we can do much more than what we currently have. Professional physics engines are already pretty optimised, they just lack some features that could be implemented with a lot of time and effort. Now that dedicated physics engines are a reality, I think we'll see more and more features appearing in the games. Already, the features in UT2004, Half Life 2 and Flat Out (full on rigid body dynamics, with hundreds of objects interacting, breakable objects, ragdolls) are pretty common in games, and will be the basic stuff for the next generation. But even that requires a LOT of know-how and work. Maybe in a couple of centuries, they'll teach rigid body dynamics, constraints, lagrangian dynamics to 4th graders :)
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
