El mié, 19-09-2007 a las 09:49 -0400, Kenneth Lerman escribió: > 100,000 iterations sounds like way too many. I suggst that you use a > variable number of iterations based on some convergence criterion. > > Also, if you save the answer from the previously calculated forward > kinematics and use that as the initial value of the new calculation you > will be able to save a considerable amount of time. That takes advantage of > the fact that the motion at each iteration is a relatively small distance.
Thanks Ken, I'm aware of this. I'm running 100,000 times the forward kinematics subroutine, to have an stimation of the average time of subrutine execution. Each call has about 3 to 4 Newton Raphson Iterations. If the move is too short then is posible to have only 2 iterations. Thanks Javier > Ken > > [EMAIL PROTECTED] > Mark Kenny Products Company, LLC > 55 Main Street Voice: (888)ISO-SEVO (888)476-7386 > Newtown, CT 06470 Fax: (203)426-9138 > http://www.MarkKenny.com > > > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] Behalf Of jros > Sent: Monday, September 17, 2007 12:56 PM > To: Enhanced Machine Controller (EMC) > Subject: [Emc-users] Debugging parallel kinematics: how to printf in > dmesg? > > > Hello!. > > We are implementing a 6 dgf kinematics for a parallel manipulator. > > (see http://imac/parallel/images/argazkiak/img_4336.jpg for a photograph > of the machine without the head) > > We are using as a template genhexkins.c file so that compiling and > launching is not a problem. > > We have provided new implementations for forward and inverse kinematics. > > And we are continuously hanging ./scripts/emc hexapod-sim. Well, the > computers gets frozen, and we have to reset it!!!!!. > > I've compiled the kinematics outside emc providing a main, and it seems > that we have done a correct programing. > > Forwardkinematics requires a newton raphson iteration, much longer to > compute compared with Inverse kinematics. > > A 9 newton raphson iteration calculation requires on a 100000 iteration > average a calculation time of 0.000797 secs > in our computer. > > As emc hangs we've tried to use trivkins to replace the forward and > inverse kins in genhexkins.c and it works ok, it does not hangs. > > As a last try, we have forced our kinematics to output the trivkins > kinematics, but forcing it to internaly deal with all the computations > of our model (we are forcing 9 newton-raphson iterations), in order to > see if it is a "too time consuming subrutine" problem. > > It works, it seems it works slower that with the surplus of > computations. > > But we are eye blinded to to see what is happening in fact within our > subrutine, how many iterations? output values? and input values?. > > Being the kinematics a real time module, I presume that printf debuging > will not work, more if the aplication freezes the computer. > > I presume the only posibility is to use dmesg, but I don't know how to > send output so that I can see what is happening with dmesg > > What can I do?, or where can I read about dealing with this type of real > time subrutine debugging?. > > I think I can try to debug using dmesg approach, using a working > kinematics -so the computer does not get freezed- (as the trivkins > approach mentioned before) to see what are the subrutines called and the > values of the variables passed to them, and hopefully discover what is > going on. > > Thanks in advance > > Javier Ros & Aitor Plaza > > > > > > > ------------------------------------------------------------------------- > This SF.net email is sponsored by: Microsoft > Defy all challenges. Microsoft(R) Visual Studio 2005. > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ > _______________________________________________ > Emc-users mailing list > Emc-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/emc-users ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
