Looking Forward - What's Next?

Ubisoft Montreal's Assassin's Creed

It took LL years to implement the Havok 4.0 upgrade of their chosen physics engine.

Havok, owned by Intel, has been delivering tools and SDKs for more then physics for years. With LL being so Avatar driven, and animation and movement being the a core differentiator from all other view driven (mouselook) 3D environments, one would think that LL's R&D department must be looking at their laggy, choppy animation piplines. Recently I ran across an interview (clipped below) with Havok's Jeff Yates @ Gamasutra.com. Give it a read and just think what the next 5 years may bring.

Havok is mainly known for its physics, but your animation and behavior components have been in development for a few years now, right?

Jeff: Yes. We had the coming-out launch of our animation product, Havok Animation, at Game Developers Conference (GDC) 2005. Since then we have continued to mature the product, and it is now in use in dozens of games. We've developed a complete content pipeline to help game developers get the animation assets in the game as effectively as possible.

For example, to deal with the natural memory limitations on many game systems, Havok Content Tools gather all the animation tracks in 3ds Max, Maya, and XSI, and compress them in a way that can be efficiently decompressed on a range of game consoles from the PS3 to the Wii. Compressing animation is a bit like compressing video: it relies on clever off-line analysis and packing that enables fast and relatively high-quality decompression in the game.

These compression settings are all controllable and previewable via the Havok Content Tools that install directly into popular 3D modeling tools. So you can get a sense of how much compression you are getting up front.

Havok Animation also offers various inverse kinematics (IK) solvers, with runtimes appropriate for game engines., We have Foot IK, which can adjust and climb over uneven terrain, as well as Hand IK, which can dynamically assess specific grab orientations for properly picking up objects and holding onto environmental elements.

Then there's what we call ragdoll mapping, which is basically runtime motion mapping between a simplified physical representation of the character, and its traditional "rig" or skeleton. For example, say you have a simple ragdoll specification with only a few spine bones (not a ton, because you don't need them for the physics), and also the higher-resolution bone system -- between 50 and 100 bones -- which is what animators usually work with.

With real-time ragdoll mapping, we can let the physics drive the animation, the animation drive the physics, or some combination of the two. We do this to achieve full-body IK-like effects -- with the added benefit that limbs do not pass through themselves or the body, nor through any other objects they interact with. It's the best of both worlds.