New paper about impulse-based dynamic simulation

Please don't post Bullet support questions here, use the above forums instead.
Jan Bender
Posts: 111
Joined: Fri Sep 08, 2006 1:26 pm
Location: Germany
Contact:

Post by Jan Bender »

yes however if i see a sparse LU decomposition i think that the solver i designed for large n .
But the k can be very big due to other reasons. For example, if you use the Lagrange multipliers, you have to compute continuous constraint forces. When integrating the body states e.g. with Runge Kutta 4, you have to evaluate these forces for different time step sizes. By using the impulse-based approach you have to compute the impulses just once per time step. So the constant k will be probably smaller.

We have a good implementation of Baraffs method and the results I mentioned were surprising but they were made with a fair test. I regret that I don't have an implementation of the Featherstone algorithm. It would be interesting to compare all methods.

Jan
Jan Bender
Posts: 111
Joined: Fri Sep 08, 2006 1:26 pm
Location: Germany
Contact:

Post by Jan Bender »

After the discussion about the complexity of my impulse-based method I tried to formulate the system of linear equations (SLE) for the impulses in an equivalent way as Baraff did for the Lagrange multipliers. And in fact I found an equivalent formulation of the problem. So I was able to solve the SLE in the same way as Baraff. This works in linear time (and requires O(n) space). :D

By computing the impulses in linear time even the simulation of the tree with 255 joints was faster than real-time on my 3.4 GHz Pentium. So it was faster than using PARDISO.

I probably will write a paper about the linear-time dynamics with my impulse-based method. You just have to wait for it :wink:

Jan
Post Reply