Is it too much to hope that someday some of the software at the heart of Pixar's physics simulation systems could be released?
Oh man, that would be awesome. When I went to the legendary SIGGRAPH 2001 Physically Based Modeling course with David Baraff and the late Andy Witkin, they discussed all the ways we could avoid having our numeric solvers "blow up" (basically, when numerical values change too quickly to figure out how to reasonably estimate them). Then Andy somewhat sheepishly mentioned that they don't actually deal with these problems, since their team at Pixar created a direct solver they actually used instead, which solved the many necessary equations symbolically instead of numerically.
This was 2001, so few in attendance had the ability to go off and write an effective direct solver for their project even if they had the right background to do it. So we made do with numeric solvers and patched over occasional blow-ups. That's a picture of how things go. It's not surprising that the pipeline in Emeryville was far more advanced, seeing how Witkin and Baraff created the Maya physics engine as a warm-up for their Pixar system. By now, there are better solvers in many physics systems... yet how much more insanely awesome must Pixar's simulation software be by now?
It's easy to see that this kind of software is highly strategic — Pixar hired world experts to work on this and it must be extremely well-seasoned code, with countless tricky "gotchas" compensated for. It's not hard to see how that gives them a clear competitive advantage.
On the other hand, the potential is enormous. This system is the masterwork, even the life's work, of unbelievably talented people. Erin Catto's work and generosity with Box2D has shown the massive innovation and even cultural value of code like this. How many more amazing things could be created with a simulation engine of the quality Pixar is sitting on?
So although I'm not holding my breath, it would be amazing to see even a small part of their physics simulation in a future open source release. Is that realistic? I have no idea.
Oh man, that would be awesome. When I went to the legendary SIGGRAPH 2001 Physically Based Modeling course with David Baraff and the late Andy Witkin, they discussed all the ways we could avoid having our numeric solvers "blow up" (basically, when numerical values change too quickly to figure out how to reasonably estimate them). Then Andy somewhat sheepishly mentioned that they don't actually deal with these problems, since their team at Pixar created a direct solver they actually used instead, which solved the many necessary equations symbolically instead of numerically.
This was 2001, so few in attendance had the ability to go off and write an effective direct solver for their project even if they had the right background to do it. So we made do with numeric solvers and patched over occasional blow-ups. That's a picture of how things go. It's not surprising that the pipeline in Emeryville was far more advanced, seeing how Witkin and Baraff created the Maya physics engine as a warm-up for their Pixar system. By now, there are better solvers in many physics systems... yet how much more insanely awesome must Pixar's simulation software be by now?
It's easy to see that this kind of software is highly strategic — Pixar hired world experts to work on this and it must be extremely well-seasoned code, with countless tricky "gotchas" compensated for. It's not hard to see how that gives them a clear competitive advantage.
On the other hand, the potential is enormous. This system is the masterwork, even the life's work, of unbelievably talented people. Erin Catto's work and generosity with Box2D has shown the massive innovation and even cultural value of code like this. How many more amazing things could be created with a simulation engine of the quality Pixar is sitting on?
So although I'm not holding my breath, it would be amazing to see even a small part of their physics simulation in a future open source release. Is that realistic? I have no idea.