Menu
▾
▴
Re: [Dvc-devel] Questions about additional DVC capabilities.
|
From: Joe H. <hay...@gm...> - 2008-06-24 14:42:11
|
Thanks Steve, This does help. Case #2 is what I'm looking for. I've not yet looked into the joint code. Looks like its time... I appreciate your support. Joe On Tue, Jun 24, 2008 at 10:19 AM, Steve Berard <sb...@cs...> wrote: > Joe, > Depending on what you are asking, we might be able to do this. > > Given two bodies, body1 and body2 and two points on the bodies, p1 and p2 > respectively: > > 1) Constrain the points p1 and p2 to lie on top of each other and constrain > the angular velocity of body2 to zero. > > 2) Constrain the points p1 and p2 to lie on top of each other and constrain > the angular velocities of body1 and body2 to be equal. > > If you want case one, we can currently do this. If case 2, we do not have > a joint that will do this, however, it would be very easy to write. > > For 1), you can use the DegreeOfFreedom constraint to remove the theta > degree of freedom from the body. There is an example of this in > examples/ropeExample > > For 2), we would have to write a new bilateral constraint very similar to > the Revolute Joint. The difference would be that the new constraint would > also constrain the angular velocities, and the extra terms in the Jacobian > matrices would be > m_Jacobians[0](2,2) = 1.0; > m_Jacobians[1](2,2) = -1.0; > The stabilization term would be the difference in the original angle > between the bodies, to the current angle. > > Let me know if this helps and what case you are interested in. > > Thanks, > > --Steve > > Joe Hays wrote: > >> Another quick question. >> >> It appears that there is not 'fixed' joint. Am I correct? I'd like to make >> a 'dumbell' like body that has 'point' masses at the extremes of a massless >> rod. I can join the 'kinematic' rod to the 'point' masses on the ends with a >> revolute joint but I'd actually like to remove the degree of freedom. Its >> not the end of the world if there's not solution... I can formulate this >> differently but if you have any suggestions of how to 'fix' a dynamic body >> to a kinematic body somehow I'd appreciate the advice. >> >> Thx. >> >> Joe >> >> On Tue, Jun 10, 2008 at 7:30 AM, Joe Hays <hay...@gm... <mailto: >> hay...@gm...>> wrote: >> >> Steve, >> >> Sounds great. I appreciate your help! I'll play around with the >> new damping support in the next couple days. I'm also not apposed >> to investigating how to add the new 'energetic' bodies but that >> may not come until later in the summer as I'm quite sure this is a >> more involved task. >> >> I'll be in touch. >> >> Joe >> >> >> On Mon, Jun 9, 2008 at 7:33 PM, Steve Berard <sb...@cs... >> <mailto:sb...@cs...>> wrote: >> >> Joe, >> To follow up the first email, damping of the joints was not as >> easy to do as I initially thought it would have been because >> of some technical issues. I spent the time to add the damping >> terms to the bilateral constraints directly, so this should >> hopefully be easy to use now. I just committed the changes >> and you can see the pendulum examples Pat committed earlier >> for an example of how to add damping to joints. It should be >> as simple as now adding >> <pref name='dampingCoef' type='double' value='10'/> >> inside the joint description in the xml file. >> >> I like the sound of adding a new "energetic" body type. Feel >> free to do this and we'll commit it to the code :) >> >> --Steve >> >> Joe Hays wrote: >> >> Thanks Steve. I appreciate the support. >> >> *"Right now dvc has no notion of a spring or damper." * >> Not that I'm sufficiently familiar with the code nor have >> I looked into the feasibility of this, but, it seems if >> one were to add a new body type, say 'energetic', along >> with 'dynamical', 'kinematic', and 'obstacle' one could >> define elements such as a spring and/or damper such that >> their wretches on the system are implicitly handled by the >> solver... Since we're dreaming... it would be good if this >> was a plug-in interface such that new body types could be >> added to the system in such a manner that they were >> implicitly handled by the solver. I'm sure this falls into >> the long list of 'wouldn't it be nice if...". Anyway, >> you've told me what I needed to know and will hopefully >> play with it a bit to try it out. >> >> *"If no value is specified by the user, then the global >> value is used."* >> Seems like a reasonable solution. This gives the ability >> to specifically define the coeffs. for specific contacts >> of interest and the ease of assuming a global setting for >> the rest. >> >> Joe >> >> >> On Mon, Jun 9, 2008 at 12:56 PM, Steve Berard >> <sb...@cs... <mailto:sb...@cs...> >> <mailto:sb...@cs... <mailto:sb...@cs...>>> >> >> wrote: >> >> Hi Joe, >> You have raised some very good questions. I'll do my >> best to >> answer them. >> >> 1) Right now dvc has no notion of a spring or damper. >> If you wish >> to attach a spring-damper to a body, the only way is >> through the >> forceController class. What this means is that you the >> user will >> be forced to compute the spring-damper forces as a >> function of the >> body's state. This is actually how the mouse >> interaction works. >> We attach one end of a virtual spring-damper to the >> body and the >> other to the mouse pointer. This spring force is then >> added to >> the timestepper code through the applied force vector. >> >> >> 2) Pat is right, we currently do not support joint >> dissipation >> directly through the xml file (we should and if I have >> time I'll >> try to hack that in). I don't know of a way to add Coulomb >> friction to a joint if the joint is abstracted as a >> bilateral >> constraint. Viscous friction, however, is possible. >> The way you >> can do it is by attaching a forceController to the body >> with the >> joint, and computing the damping force. Here is what I >> would do >> for revolute joint damping. >> a) attach a forceController to the body with a >> revolute joint >> b) In the GetForce function of the ForceController >> i) get the angular velocity of the body attached to the >> revolute joint (vel[2]) >> ii) set outForce[2] = -C*vel[2] >> where C is your damping coefficient. >> Force is a bad name for this function, it really should >> be called >> getWrench since the 3 dimensional value you are >> returning is the >> force and moment. >> >> For dry friction in a joint, I think you would have to >> model the >> joint geometrically and drop the bilateral >> representation. If you >> know of a way to add dry friction into the bilateral joint >> representation I would love to hear it, because (to the >> best of my >> knowledge) this would be something that no one else is >> doing. >> >> 3) This is something that Pat and I have talked about >> in the past >> without an ideal solution. The best idea we currently have >> (un-implemented of course) is to keep the global value >> (like the >> current set up), but to allow the user to specify a >> specific value >> at a per-body-pair level. If no value is specified by >> the user, >> then the global value is used. If this would be >> sufficient for >> what you need let us know, because that shouldn't be >> very hard to >> do. If you can think of a better solution, also please >> let me know. >> >> I hope this answers all of your questions, feel free to >> email me >> back if anything is unclear or if you have more >> questions. I have >> also copied this message to the dvc-devel mailing list >> so a public >> copy is archived. >> >> Thanks, >> >> --Steve >> >> >> >> Joe Hays wrote: >> >> Steve, >> >> I've learned quite a bit in the last two weeks or >> so about >> DVC. With Pat's help I've successfully made a >> double pendulum, >> added a feedback controller to it and observed >> closed-loop >> behavior (even though my current controller design >> sucks... I >> know the loop is closed and that's what was >> important to me.) >> >> The following questions are outstanding in my mind. >> (Pat has >> indicated that you're the right person to elaborate >> on these >> topics. If you have any light you can shed on these >> topics I'd >> appreciate it.) >> >> *How to make the classic "mass-spring-damper" model?* >> From what I've seen so far DVC's unique strength is >> in solving >> the Coulomb Friction and Impact dynamics for a >> rigid-body-system. However, I've not seen how to >> create some >> of the classical models to be simulated (e.g. the >> "mass-spring-damper"). Is there a way I can define >> general >> elements such as a spring, with either linear or >> nonlinear >> constitutive law, as well as a damper (such as a >> viscous >> damper which is either linear or non-linear)? >> >> *How to add dissipation (i.e. viscous and/or >> Coulomb friction, >> preferably both) to joints?* >> Pat indicated I need to 'hack' this in as their is >> no direct >> facility for this. Can you give me some guidance on >> this? >> >> *Unique friction and restitution coefficients?* >> From what I've learned so far both the friction and >> collision >> restitution coefficients are global for the scene. >> Do you have >> plans, back door entry points, or hacks, to allow >> these to be >> defined for each body combination pair? (This is less >> important for now but will become important to me >> in the >> future...) >> >> I'm enjoying learning your system. Congratulations >> on the >> progress you've made thus far with DVC, and, thanks >> for the >> support. >> >> Joe >> >> >> >> >> >> >> |
