Articulated systems & indep. generalized coordinates

[kalesony]

Intro: When simulating (heavily) articulated systems, among all other circumstances and options to consider, one needs to decide whether to use a set of independent generalized coordinates or to go for a redundant set with explicitly stated inter-relations. This is a true statement, isn't it?

As far as I can tell (which is not that far, I'm afraid ) the solutions visible/available on the market (proprietary or open) seem to prefer the latter option (redundant set): is my observation correct?

Question: Could any of you better-informed-than-me either confirm the above or prove me wrong by providing examples? It would be of great help and will quite probably seriously influence my decision as to what I should do next .

Thanks.

--
P.S. The background for this question is that I've been quite literally forced (university grant) to stick to independent coordinates (Featherstone-based formulations) and now, after quite some time of trying to wrap my head around it and provide a working implementation, I finally decided to look around and see if I'm anywhere near to what has actually been done in this field over last the past x years (wrong order of actions, now I know).

[DannyChapman]

PhysX 3 contains articulations, which you should look at. They work very well...

[kalesony]

Danny - thanks for the speedy reply but I am not sure I understand you correctly: are you saying that PhysX enforces articulation/joint constraints using independent set of coors?

[Erwin Coumans]

Reduced coordinates methods, such as Featherstone, are being used, several games shipped with Bullet+Featherstone Articulated Body Method.

For Bullet 3.x we add the option for Featherstone as well. We received 3 separate ABM implementations, one of them is already open source:
See RBDL and https://code.google.com/p/bullet/issues/detail?id=33

[kalesony]

Erwin, and what were the reasons for using them instead of sticking to "core" Bullet? Ragdolls maybe? Still, Bullet's ragdoll sample seems to work pretty well.


-
In case you missed it - I pm-ed you about "implementing Featherstone" in Bullet.

[DannyChapman]

Danny - thanks for the speedy reply but I am not sure I understand you correctly: are you saying that PhysX enforces articulation/joint constraints using independent set of coors?

I'm saying you should look up the documentation and have a look at the api for what PhysX call Articulations, as it may give you some clues about how they implement things, if that is what you're looking for. Unfortunately, probably not a big clue, and I can't really comment on the internal workings However, the end result is very good - you can have a ragdoll that is stable for large timesteps (30FPS with a human character), exhibits very little joint separation, and (very importantly for me) provides a very stable and accurate drive model.

[kalesony]

Danny - okay, I will check this out, thanks.