Provided files and other resources

There are three folders given in the package:

1. Common contains .h and .cpp files that have Probes, EventHandlers, and the MuscleLikeCoordinateActuator that are not in the OpenSim distribution. It also has files for the Covariance Matrix Adaptation (CMA) optimizer.

2. InitialFiles has a provided model file (AshbyModelv8_twoConstraints.osim) and a controls file for a nominal standing long jump (InitialControllerParameters.sto). Use this controls file as a template to make new controls files.

3. nodal contains a .cpp file that creates the main() program to load a model and either perform an optimization or run a forward simulation with various analyses attached.

This model is based on a model previously developed by Ashby and Delp. Ashby used the model to study the role of arms in a long jump in his 2006 paper, and a more detailed description is provided in his thesis. Differences in this implementation include

• Ligament forces are modeled using CoordinateLimitForce in OpenSim.

• Two point constraints are used at the toe and the heel, rather than just the toe.

Since this problem is a non-convex, non-linear optimization, the CMA optimizer provided and used as default. For more information about how this optimizer works, please refer to its Wikipedia page or tutorial slides. The CMA wrapper provided here also has an option to enable MPI to parallelize calculations for the optimization.

CMake Configuration Options

The root directory of the package has a CMakeLists.txt file that can be used to build the project.

• ENABLE_MPI: If checked, the project will be able to use the Message Passing Interface (MPI) standard to allow parallelization of the CMA optimizer. 

• MPI_INSTALL_DIR: Used for Windows OS. Assumes that the Microsoft HPC Pack has been installed and will be used as the MPI implementation. (If using UNIX, this will be ignored and the MPICH libraries are assumed to be used).

• NameSpace: Leave blank if building from source. Use "OpenSim_" when linking to pre-built libraries.

• OPENSIM_INSTALL_DIR: Point to the install directory for OpenSim.

Building the project

Building the ALL_BUILD project will build two projects:

1. Executables - StandingLongJump_nodal: This will build StandingLongJump_nodal.exe. This executable is used for running both a single forward simulation or for performing an optimization.

2. Libraries - sljCommon: This will build a dynamic linked library (sljCommon.dll or sljCommon.so) that contains extra classes not included in the distribution of OpenSim. This includes Probes, EventHandlers, and the MuscleLikeCoordinateActuator needed for the model and optimization as well as an implementation of the CMA optimizer.

Running a single forward simulation with analyses

To run a single forward simulation with analyses, we need to have

1. The executable, StandingLongJump_nodal.exe

2. The common library, sljCommon.dll. This needs to either be in the same directory as the executable or on the system path for Windows.

3. Input model file without controls (i.e. AshbyModelv8_twoConstraints.osim)

4. Controls file (i.e. InitialControllerParameters.sto)

Options:

• -m <modelfile.osim>: Specify a model file to use. Default is AshbyModelv8_twoConstraints.osim if not specified. 

• -cf <controlsfile.sto>: Specify a controls file to use. Default is InitialControllerParameters.sto if not specified.

The executable can be run from the command prompt. As an example, you can run the executable as:

This will output the following files:

• forceReporter.sto: Output from a ForceReporter.

• fwdLog.txt: Contains information about length of simulation, integrator tolerance, and values that are used to calculate the objective function.

• PointKinematics_HEEL_*.sto: PointKinematics analysis of the heel.

• PointKinematics_TOE_*.sto: PointKinematics analysis of the toe.

Performing an optimization

To perform an optimization, we need to have

1. The executable, StandingLongJump_nodal.exe

2. The common library, sljCommon.dll. This needs to either be in the same directory as the executable or on the system path for Windows.

3. Input model file without controls (i.e. AshbyModelv8_twoConstraints.osim)

4. Controls file with the initial guess for the optimizer (i.e. InitialControllerParameters.sto)

Options:

• -m <modelfile.osim>: Specify a model file to use. Default is AshbyModelv8_twoConstraints.osim if not specified. 

• -cf <controlsfile.sto>: Specify a controls file to use. Default is InitialControllerParameters.sto if not specified.

• -opt: Flag that specifies that an optimization will be performed. This flag must be used when performing an optimization.

• -lambda <int>: CMA optimization parameter specifying number of samples per generation. Default is 100 if not specified.

• -sigma <positive real number>: CMA optimization parameter specifying constant scaling term of initial covariance matrix.

• -maxIters <int>: Optimization parameter limiting number of outer iterations (a.k.a. generations for CMA). Default is 1000 if not specified.

• -resume: If resumecmaes.dat file is present and the -opt flag is used, this flag specifies the CMA optimizer to resume from a previous optimization corresponding to the file. This will override any -sigma setting, but -lambda and -maxIters can still be specified. If resumecmaes.dat is not present, then the optimizer will begin a new optimization. If the -opt flag is not used, then the -resume flag will have no effect and a single forward simulation will be performed.

The executable can be run from the command prompt. As an example, you can run the executable as:

If the ENABLE_MPI flag was set to true in CMake and the proper MPI libraries are linked, you can use more than one thread to parallelize calculations. An example way to run this (in this example using 8 threads) is:

An example to resume a previous CMA optimization is to run the following in the command prompt:

This will output the following files specific to this project:

• ControllerParameters_gen<int>.sto: This is the controller parameters corresponding to the best objective function value for the generation number. This is set to print the 1st and every 100th generations.

• optimizedControls.sto: Controller parameters corresponding to the best objective function found.

• optimizedModel.osim: Model with the controller and control parameters corresponding to optimizedControls.sto.

• optLog.txt: Lists the options and parameters used for the optimization.

• outputlog.txt: Tab-delimited file with information about the progression of the optimizer. The columns contain: 1) generation number, 2) best objective function value, 3) sigma for the current generation, and 4) corresponding controls for the best objective function. This is printed for the 1st and every 50th generations.

• resumecmaes.dat: Generated by the CMA optimizer to allow the user to resume an optimization.

Objective Function