not accurate results from Moco Track

OpenSim Moco is a software toolkit to solve optimal control problems with musculoskeletal models defined in OpenSim using the direct collocation method.
User avatar
Mohammad RahimiGoloujeh
Posts: 14
Joined: Thu Jul 01, 2021 5:57 am

not accurate results from Moco Track

Post by Mohammad RahimiGoloujeh » Thu Sep 30, 2021 9:15 am

Hello,
I am trying to track experimental data (markers) and get the motion but compared with the inverse solution the result is far off.
I am using OpenSim 4.2 and python upgraded to 4.3 same results. however, tested the same code on a computer with OpenSim 4.1 gave me much better results.
does anyone know how I can fix this
appreciate your help

User avatar
Nicholas Bianco
Posts: 1041
Joined: Thu Oct 04, 2012 8:09 pm

Re: not accurate results from Moco Track

Post by Nicholas Bianco » Thu Sep 30, 2021 10:50 am

Hi Mohammad,

Do you mean that the muscle activations are different, or the kinematics? And how are they different?

Results from a tracking problem can be quite different from an inverse problem depending on what the tracking solution kinematics and ground reaction forces look like.

-Nick

User avatar
Mohammad RahimiGoloujeh
Posts: 14
Joined: Thu Jul 01, 2021 5:57 am

Re: not accurate results from Moco Track

Post by Mohammad RahimiGoloujeh » Wed Oct 06, 2021 7:55 am

ankle_comparison.pdf
(16.43 KiB) Downloaded 27 times
Hi,
Thank you for your reply.
let me explain more. So i plotted the results of the same code solved with opensim_moco integrated 4.2 and 4.3 and also previous version of moco when it was not integrated with opensim i believe the version for it is 0.4.
the results for tracking of moco 0.4 is about the same as inverse solution whereas the results of 4.3 and 4.2 are vastly different and inaccurate.
    i also checked the same code on a different computer same results were observed.
      i also modified the solver same results were observed. important to note that in all of the cases i received the message indicating solver succeeded.
      in terms of computation time in 4.3 and 4.2 it is much less which i think some connection between the solver and moco and/or opensim and/or python is broken. as can be seen from the results it didnt solve anything:
      Attachments
      ankle_comparison3.pdf
      (16.98 KiB) Downloaded 19 times
      Last edited by Mohammad RahimiGoloujeh on Wed Oct 06, 2021 8:38 am, edited 1 time in total.

      User avatar
      Mohammad RahimiGoloujeh
      Posts: 14
      Joined: Thu Jul 01, 2021 5:57 am

      Re: not accurate results from Moco Track

      Post by Mohammad RahimiGoloujeh » Wed Oct 06, 2021 7:58 am

      below is the code im using
      Last edited by Mohammad RahimiGoloujeh on Wed Oct 06, 2021 8:34 am, edited 1 time in total.

      User avatar
      Mohammad RahimiGoloujeh
      Posts: 14
      Joined: Thu Jul 01, 2021 5:57 am

      Re: not accurate results from Moco Track

      Post by Mohammad RahimiGoloujeh » Wed Oct 06, 2021 8:33 am

      Code: Select all

      # -------------------------------------------------------------------------- #
      # OpenSim Moco: exampleMocoTrack.py                                          #
      # -------------------------------------------------------------------------- #
      # Copyright (c) 2019 Stanford University and the Authors                     #
      #                                                                            #
      # Author(s): Nicholas Bianco                                                 #
      #                                                                            #
      # Licensed under the Apache License, Version 2.0 (the "License") you may     #
      # not use this file except in compliance with the License. You may obtain a  #
      # copy of the License at http://www.apache.org/licenses/LICENSE-2.0          #
      #                                                                            #
      # Unless required by applicable law or agreed to in writing, software        #
      # distributed under the License is distributed on an "AS IS" BASIS,          #
      # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.   #
      # See the License for the specific language governing permissions and        #
      # limitations under the License.                                             #
      # -------------------------------------------------------------------------- #
      
      # This example features two different tracking problems solved using the
      # MocoTrack tool. 
      #  - The first problem demonstrates the basic usage of the tool interface
      #    to solve a torque-driven marker tracking problem. 
      #  - The second problem shows how to customize a muscle-driven state tracking 
      #    problem using more advanced features of the tool interface.
      # 
      # See the README.txt next to this file for more information.
      
      import os
      import math
      import opensim as osim
      
      def torqueDrivenMarkerTracking():
      
          # Create and name an instance of the MocoTrack tool.
          track = osim.MocoTrack()
          track.setName("torque_driven_marker_tracking1")
      
          # Construct a ModelProcessor and add it to the tool. ModelProcessors
          # accept a base model and allow you to easily modify the model by appending
          # ModelOperators. Operations are performed in the order that they are
          # appended to the model.
          # Create the base Model by passing in the model file.
          modelProcessor = osim.ModelProcessor("standingperturbationScaled_Prescribed.osim")
          # Add ground reaction external loads in lieu of a ground-contact model.
          modelProcessor.append(osim.ModOpAddExternalLoads("Trial26_GRFFile_wrtOrigin.xml"))
          # Remove all the muscles in the model's ForceSet.
          modelProcessor.append(osim.ModOpRemoveMuscles())
          # Add CoordinateActuators to the model degrees-of-freedom. This ignores the 
          # pelvis coordinates which already have residual CoordinateActuators.
          modelProcessor.append(osim.ModOpAddReserves(250))
          track.setModel(modelProcessor)
      
          # Use this convenience function to set the MocoTrack markers reference
          # directly from a TRC file. By default, the markers data is filtered at
          # 6 Hz and if in millimeters, converted to meters.
          track.setMarkersReferenceFromTRC("Trial26_markers_rotated.trc")
      
          # There is marker data in the 'marker_trajectories.trc' associated with
          # model markers that no longer exists (i.e. markers on the arms). Set this
          # flag to avoid an exception from being thrown.
          track.set_allow_unused_references(True)
      
          # Increase the global marker tracking weight, which is the weight
          # associated with the internal MocoMarkerTrackingCost term.
          track.set_markers_global_tracking_weight(20)
      
          # Increase the tracking weights for individual markers in the data set 
          # placed on bony landmarks compared to markers located on soft tissue.
          markerWeights = osim.MocoWeightSet()
          markerWeights.cloneAndAppend(osim.MocoWeight("RASI", 20))    
          markerWeights.cloneAndAppend(osim.MocoWeight("LASI", 20))
          markerWeights.cloneAndAppend(osim.MocoWeight("SACR", 10))
          markerWeights.cloneAndAppend(osim.MocoWeight("RKNE", 10))
          markerWeights.cloneAndAppend(osim.MocoWeight("RANK", 10))
          markerWeights.cloneAndAppend(osim.MocoWeight("RHEE", 5))
          markerWeights.cloneAndAppend(osim.MocoWeight("RTOE", 2))
          track.set_markers_weight_set(markerWeights)
          ##
          
      
          # Initial time, final time, and mesh interval. The number of mesh points
          # used to discretize the problem is computed internally using these values.
          track.set_initial_time(0.5)
          track.set_final_time(2.5)
          track.set_mesh_interval(0.005)
          
          ######## adding control effort as a goal
          #effort = osim.MocoControlGoal()
          #effort.setName("effort")
          #effort.setWeight(5)
          #problem.addGoal(effort)
          #################################################
          
          #adding torso orientation goal:
          #torsoGoal= osim.MocoOrientationTrackingGoal('torsoGoal',10)
          #torsoTable = osim.TableProcessor('refQ_torso_orientation.sto')
          #torsoGoal.setStatesReference(torsoTable)
          
          #torsoGoal.setRotationReferenceFile('torso_orientation.sto');
          #paths = []
          #paths.append('/bodyset/torso')
          #torsoGoal.setFramePaths(paths)
          #problem.addGoal(torsoGoal);
          ###################################################
          
          #####################################################
      
          # Solve! Use track.solve() to skip visualizing.
          study=track.initialize()
          problem=study.updProblem()
          solver = study.initCasADiSolver()
          solver.set_num_mesh_intervals(80);
          solver.set_verbosity(2);
          solver.set_optim_solver("ipopt");
          solver.set_optim_convergence_tolerance(1e-4);
          solver.set_optim_constraint_tolerance(1e-4);
          solver.set_optim_max_iterations(2000)
          ##solver.set_optim_convergence_tolerance(1e-4)
          a=solver.get_optim_ipopt_print_level()
          print("solver.get_optim_ipopt_print_level()",a)
          c=solver.get_optim_max_iterations()
          print("solver.get_optim_max_iterations",c)
          d=solver.get_optim_convergence_tolerance()
          print("solver.get_optim_convergence_tolerance()",d)
          
          solution = study.solve()
          
          b=solver.get_optim_ipopt_print_level()
          print("solver.get_optim_ipopt_print_level",b)
          
      
          
          #solution = track.solve()
          solution.unseal()
          #track.visualize(solution)
          solution.write('track_balance_perturbed_control_No_pelvisPosition_moco_0_4.sto')
          b=solution.getFinalTime()
          c=solution.getNumIterations()
          d=solution.getSolverDuration()
          e=solution.getStatus()
          a=solution.getParametersMat()
          print("finaltime", b)
          print("iteration", c)
          print("duration", d)
          print("statuse=", e)
          print("getParametersMat",a)
          #solution.write('track_balance_perturbed_control.sto')
          
          '''
          osim.STOFileAdapter.write(solution.exportToStatesTable(),
                            "moco_track_reactionGoal_Muscle_states.sto")
          osim.STOFileAdapter.write(solution.exportToControlsTable(),
                            "moco_track_reactionGoal_controls.sto")
          '''
          
          #JR_paths = []
          #JR_paths.append('.*reaction_on_child')
          #statesTable = solution.exportToStatesTable();
          #controlsTable = solution.exportToControlsTable();
          #model=osim.Model("subject_walk_armless.osim")
         # JR_outputs_table = osim.analyzeSpatialVec(model, statesTable, controlsTable, JR_paths).flatten()
          #JR_outputs_table.write('reaction.sto')
          
      ################# running the solution  ##################################################  
      
      torqueDrivenMarkerTracking()

      User avatar
      Nicholas Bianco
      Posts: 1041
      Joined: Thu Oct 04, 2012 8:09 pm

      Re: not accurate results from Moco Track

      Post by Nicholas Bianco » Thu Oct 07, 2021 12:05 pm

      Hi Mohammad,

      I'm still not sure what you mean by "vastly different and inaccurate" when referring to the solutions using 4.2 and 4.3. The plot you included seems to only compare solutions when using Moco 0.4. Various changes have been implemented in Moco since we merged the codebase into OpenSim. As a result, some of the examples may behave slightly differently. For example, Moco 0.4 had an issue where muscle paths were not being converted correctly when replacing muscles with the DeGrooteFregly2016Muscle; that has been fixed in 4.2 and 4.3 and might change the results in the MocoTrack examples.

      If you're looking for better tracking of the ankle angle, you might need to just increase the tracking weight for that joint in the MocoStateTrackingGoal.

      -Nick

      User avatar
      Mohammad RahimiGoloujeh
      Posts: 14
      Joined: Thu Jul 01, 2021 5:57 am

      Re: not accurate results from Moco Track

      Post by Mohammad RahimiGoloujeh » Fri Oct 08, 2021 11:15 am

      Hi ,
      thanks again for your help and time.
      So I included two plot please also take a look at the other one that i compared same code solved by different versions
      so when I mentioned track only I commented out other goals. it is just tracking and i think the weight of the tracking doesn't make any difference

      User avatar
      Nicholas Bianco
      Posts: 1041
      Joined: Thu Oct 04, 2012 8:09 pm

      Re: not accurate results from Moco Track

      Post by Nicholas Bianco » Fri Oct 08, 2021 12:57 pm

      Hi Mohammad,

      Sorry, I didn't see the other file. The ankle angle for 4.3 seems fixed at a constant value. Are you sure the ankle coordinate isn't locked?

      User avatar
      Mohammad RahimiGoloujeh
      Posts: 14
      Joined: Thu Jul 01, 2021 5:57 am

      Re: not accurate results from Moco Track

      Post by Mohammad RahimiGoloujeh » Mon Oct 11, 2021 9:47 am

      Hi Nick,
      Thank you so much for your time and help.
      yes I'm sure that is not locked. I used the same everything only different versions of moco.
      As i told you before when i simulate it with newer version it takes very little time to give me the results which i think the reason is it doesnt solve it at all. on the other hand with moco 0.4 it takes like 10 times more to solve.
      this made me think maybe some connections with the solver is broken. or some incompatible package is the reason.
      we tried on a different computer with opensim 4.2 result were not correct either on that.
      i uploaded the model im using for your referance
      Attachments
      standingperturbationScaled_Prescribed.osim
      (169.14 KiB) Downloaded 11 times

      User avatar
      Mohammad RahimiGoloujeh
      Posts: 14
      Joined: Thu Jul 01, 2021 5:57 am

      Re: not accurate results from Moco Track

      Post by Mohammad RahimiGoloujeh » Mon Oct 11, 2021 10:02 am

      one other thing I wanted to mention is that. the model is scaled and the platform motion is prescribed using an older version of OpenSim (4.1).
      however, I don't know if it's something important

      POST REPLY