Open Dynamics Engine (ODE)

☞ http://opende.sourceforge.net

ODE is a free, industrial quality library for simulating articulated rigid body dynamics - for example ground vehicles, legged creatures, and moving objects in VR environments. It is fast, flexible, robust and platform independent, with advanced joints, contact with friction, and built-in collision detection.


☞ http://opende.sourceforge.net/junk.html

☞ The competition

Why does the world need another dynamics engine? The answer to this question will be written later. Meanwhile I list below some other rigid body libraries. I compare them on features like speed and stability, but this is not really meaningful as each library is best suited to a particular niche. The analysis of which library is best for a particular application is much more complicated than a simple comparison of features.

• Aero
  Free. Intended for computer graphics. Has a built in 3D editor. Penalty method. Medium speed O(n). Low accuracy and stability.
  
• DynaMechs
  Free. Intended for robotics simulation. Featherstone's method. Very fast O(n). Reasonable accuracy and stability (pick your own integrator).
  
• Havok Physics SDK
  A commercial game development library. Has more goodies than MathEngine's library, like a 3DS Max plugin. Lagrange multiplier method. Medium speed O(n³).
  
• Ipion
  A commercial game development library. Ipion was bought by Havok in July 2000.
  
• MathEngine
  A commercial game development library. I wrote the core for this. Lagrange multiplier method. High speed O(n³). Low (1st order) accuracy but high stability - thus much more suitable for game applications than engineering applications.
  
• Brian Mirtich and James Kuffner's Multibody package. Free. Featherstone's method. Fast O(n). Get this from James Kuffner's software page.

Some comments on the different simulation methods:

• Penalty methods are "springs and damper" techniques. Very fast per iteration, but must be used with small time steps as they are prone to stability problems, so in the end they're not very fast at all. Do not support hard constraints, and have difficulty simulating articulated systems.
  
• Lagrange multiplier (sometimes called Cartesian) techniques, allow flexible management of hard constraints, but can be slow for large articulated systems.
  
• Featherstone based methods are very fast, but only allow tree structured systems and do not support hard contact constraints.
  
• Impulse methods are based on the work of Brian Mirtich, they are very fast but have poor support for articulated systems.