Hi everyone,
I'm trying to simulate kneeling in opensim. I'm using Gait2354 model with slight modifications(added left and right patella on it). My trial time is 30 sec. When I run IK step, it goes smoothly for couple of seconds and then I get error message:
SimTK Exception thrown at InteriorPointOptimizer.cpp:262:
Optimizer failed: Ipopt: Maximum iterations exceeded (status -1)
OPTIMIZATION FAILED...
after that it takes long time to process each frame and this message keeps repeating. Can someone help me understand why this is happening and how can I fix it?
Thanks in advance,
Mohan
Problem in IK step
-
Rudolf Klemetti
- Posts: 48
- Joined: Thu Jul 15, 2010 3:09 pm
RE: Problem in IK step
Hi Mohan -
So, IK tool tries to minimize differences between model and experimental kinematics. It performs many calculations to see which solution is the best. Each calculations is called iteration. If experimental marker data is not good or if scaling of the model is not good, IK has to take lots of iterations to match the experimental and model markers. There is a certain limit for maximum number of iterations, set by OpenSim team. So, in your case IK just didn't find a solution before maximum number of iteration was exceeded. And when IK runs slowly, many iterations are taken, but a solutions is found.
To avoid that and improve your analysis, you can try to adjust markers weights in IK tool, or go back to scale tool to see if the model could be scaled better. Reduce weights of markers that you "trust" least, and vice versa.
Rudi
So, IK tool tries to minimize differences between model and experimental kinematics. It performs many calculations to see which solution is the best. Each calculations is called iteration. If experimental marker data is not good or if scaling of the model is not good, IK has to take lots of iterations to match the experimental and model markers. There is a certain limit for maximum number of iterations, set by OpenSim team. So, in your case IK just didn't find a solution before maximum number of iteration was exceeded. And when IK runs slowly, many iterations are taken, but a solutions is found.
To avoid that and improve your analysis, you can try to adjust markers weights in IK tool, or go back to scale tool to see if the model could be scaled better. Reduce weights of markers that you "trust" least, and vice versa.
Rudi
-
Mohan Krishna Kallakanti
- Posts: 3
- Joined: Wed May 26, 2010 10:28 pm
RE: Problem in IK step
Hi Rudi,
Thank you very much for the explanation.
I will try scaling it again. I did get some IK results when I let program run for like 10 hrs. So now that I got IK step cleared, can I use that data for solving further steps like CMC?
I also wanted to know if modifying (like I did which I mentioned earlier) the model would in any way make a difference in time for solving?
Thanks,
Mohan
Thank you very much for the explanation.
I will try scaling it again. I did get some IK results when I let program run for like 10 hrs. So now that I got IK step cleared, can I use that data for solving further steps like CMC?
I also wanted to know if modifying (like I did which I mentioned earlier) the model would in any way make a difference in time for solving?
Thanks,
Mohan
-
Rudolf Klemetti
- Posts: 48
- Joined: Thu Jul 15, 2010 3:09 pm
RE: Problem in IK step
Hi Mohan,
Sounds like your IK is running very, very slowly. With my laptop, getting IK solution for 30 s trial would take a few minutes. Either your computer is slow or you have serious problems with your data. However, don't worry too much about time per se. Rather look at markers errors that IK tool reports. It provides a message for every frame, providing max and RMS errors. According to "best practices" of the OpenSim team, to get "good" simulations, RMS error should be less than 1 cm, and max error 2 cm. So, those are your goals you should try to pursue. Certainly it is possible to run CMC with greater errors but question is always that how realistic are they. The worse your experimental data is, the more difficult it is run CMC, and at some point it is impossible. I don't know if you are close to impossible, you need to try and see. Probably, you should be able to run at least RRA, if not CMC. If IK takes 10 hours, CMC might last "forever" though.
I wouldn't expect that modifications to the model make a big difference in time, as far as you have done them correctly. Have you run IK with the unmodified model?
Well, good luck!
Rudi
Sounds like your IK is running very, very slowly. With my laptop, getting IK solution for 30 s trial would take a few minutes. Either your computer is slow or you have serious problems with your data. However, don't worry too much about time per se. Rather look at markers errors that IK tool reports. It provides a message for every frame, providing max and RMS errors. According to "best practices" of the OpenSim team, to get "good" simulations, RMS error should be less than 1 cm, and max error 2 cm. So, those are your goals you should try to pursue. Certainly it is possible to run CMC with greater errors but question is always that how realistic are they. The worse your experimental data is, the more difficult it is run CMC, and at some point it is impossible. I don't know if you are close to impossible, you need to try and see. Probably, you should be able to run at least RRA, if not CMC. If IK takes 10 hours, CMC might last "forever" though.
I wouldn't expect that modifications to the model make a big difference in time, as far as you have done them correctly. Have you run IK with the unmodified model?
Well, good luck!
Rudi