This repository is linked to "On the design of simulation-assisted human-centered quasi-stiffness actuator for ankle orthosis" article (https://doi.org/10.3390/electronics13214164)
It contains all the code and data elements mandatory to set up the Human-Orthosis interaction simulation illustrated in the article.
This article, among other things, describes the design and its application on a example of an early framework focused on several simulation software, enabling the study of a MoCap recorded task performed by a human user (walking, picking up an object from the floor, climbing stairs or slopes ...) with or without external assistance provided by human-machine devices such as orthoses or exoskeletons. This framework was used to help design personnalized assistances for several simulated subjects while performing a load-carrying walking task on a flat ground as several speeds. Designs performances was then evaluated regarding muscular joint torque savings.
This framework mainly relies on both OpenSim 4.4 software - an open free software offering tools to run biomechanics analysis - and Matlab 2019 and earlier version. Additionnal GRFM estimation function need :
- Symbolic Matlab Toolbox
- Optimization Toolbox
- Parallel Computing Toolbox
- DSP System Toolbox or Signal Processing Toolbox
- Robotics System Toolbox
This early framework proposes different features :
-
The use of OpenSim tools such as "Scaling", "Inverse Kinematics", "Inverse Dynamics", "Residual Reduction Algorithm" and "Computed Muscle Control". This tool piepline enables, from motion capture data augmented with Ground Reaction Forces and Moments (GRFM), to successively determine joint kinematics from the task recording, calculate the joint torques required to generate body trajectories, and finally trace back to the muscle activation of the simulated musculoskeletal model that allows for the completion of this task. Model modification allows the designer to add extra elements on top of the body model, and the large scope of output data obtained from these tools are usefull to compute different performance metrics.
-
In the case motion capture data are not fitted with GRFM data, or to simulate the wearing of additional load absent from the task recording, a MatLab toolkit named "CusToM" presented in [Muller et al. 2019]1 allows for the prediction of GRFM from the kinematic data and the characteristics of the simulated user.
-
A code snippet is also present to extract the quasi-stiffness cycle (AQS cycles in the article) of a joint to define its torque response when generating a trajectory during the requested task.
-
Moreover, this framework also proposes a Simulink blocks diagram pattern to help define and simulate the actuation strategy of a torque-generating device assistance. It is possible to parameterize the number of active elements such as motors, or passive elements such as springs and their operating ranges and characteristics. A state machine also allows for the simulation of activation sequences of actuators to simulate complex mechanism.
-
At last, a code snippet is provided to define and calculate performance metrics of the assistance, as well as different graphical displays.
- First step is to install the latest version of OpenSim (at least 4.0) and MatLab (at least 2019).
- If you want to compute GRF prediction since you lack this data, make sure your MatLab installation fulfil the requested toolbox described above.
- I recommand also to install Mokka software to preview, label MoCap data and translate it into other formats (as an exemple, CusToM standard is .c3d, whereas OpenSim is .trc)
- Since you get all those 3 software, paste the content of the Matlab_tools/CusToM_master/Models folder of this repo into your CusToM/Models installation directory. This will allow you to use the dedicated Gait2354 model and markers adapted from OpenSim to CusToM.
Here is an overview of the HJOSET workflow to study a motion, with and without any extra device assistance.
PDF version is also available here : HJOSET_workflow.pdf
- First thing is to gather your MoCap data, idealy with ground reaction forces and moments (GRFM).
- If GRFM is missing, run the CusToM toolbox. Once your model setting is done, run the pipeline till "GRF estimation" to get the missing data according to the provided kinematics. Instructions to start and run this toolbox can be found in the "exmeples" folder of CusToM. Follow ex.1 for setup / launch / GRF estimation.
- CusToM also provide tools to add extra load onto your model to simulate load-carrying task.
- In [Mokadim et al. 2024], a methodology is provided to estimate an extra-load carrying user GRFM based on the recorded no extra-load user GRFM while ignoring inconsistencies between OpenSim and CusToM model definitions.
For every OpenSim tool mentionned earlier, default config files are provided with OpenSim exemples. You may want to edit them if you want to loop steps faster and easier. Check those available in the "Gait2354" model folder.
- First, choose a neuromuscular model to scale according to your needs.
- Get a static pose of the subject to run the Scale tool to fit the OpenSim model with the MoCap one. Make sure to fit the markers from your static pose to those of the OpenSim model.
- Once done, run the Inverse Kinematics (IK) tool.
For the next steps, you will need the GRFM data matching the task kinematics.
- You may be interested in Inverse Dynamics right after IK. Else, run the Residual Reduction Algorithm (RRA) tool to refine your model and your kinematics for the next step.
Hint : You may run this tool twice, first to get the COM-corrected model, then the second iteration applied onto the newly obtained model to get the matching corrected kinematics that minimize the residual.
- Finally, run the Computed Muscle Control (CMC) tool to obtain the muscle excitation signals, joint positions, velocities and torques.
- Since your subject is executing a cyclic motion, such as walking, running, climbing stairs up or down and so on, you can define a joint quasi-stiffness cycle (QSC)
- First thing is to plot a joint torque response against the angular position. If you get a close-loop plot, here is a QSC cycle you can use to define your device assistance profil.
- If not, in case your MoCap data are not available on a whole task cycle, you can filter/merge data (use Matlab or a spreadsheet with graphs for that ) until you get a QSC.
- Once you are done with obtaining the QSC, you can use it either to loop define and test passives actuators stiffness like springs, or use the cycle as an active actuator controler input such as motor. Use the Simulink project template to declare your actuating strategy and global mechanism to simulate the torque response of your device. The default Simulink project blocks are divided into 2 sections : main actuators are passives, and active actuators are implemented to supply the missing torque (up to a specific limit) between passive actuators and QSC reference. One if free to implement any other stategy, or just bypass any unwanted feature.
Warning As expected in the Simulink project, QSC torque component should be expressed in Nm/deg/Kg and angular position expressed in deg. This allow the designer to scale its assistance based on the amount of body mass the device should relieve. A “weight assistance” gain block is provided for this purpose.
- After that, you can check the torque generation split between scaled active and passive actuators to better observe the device inner torque generation.
- Once the torque error between your ideal unscaled device assitance and the subject's joint QSC is low enough, you can export the ouput control signal as a device controler for OpenSim CMC tool into a .sto file.
The authors of [Mokadim et al.2024] wish to compare the muscle activity of a simulated user walking without additional load with that of a user wearing a backpack and an ankle-foot orthosis on both legs. As in the modified Gait10DOF18Musc models wearing orthosis provided in the repository, it is possible to use the OpenSim "Bushing Forces" to add compliant links between the device(s) and the body model for a more realistic representation.
- Modify your model by adding an additional load and/or device by editing the .osim file.
- The next step consists in re-runing OpenSim RRA and CMC tools with an augmented model based on the task and device you wish to test.
- One can edit the mass and inertia properties of the existing ankle-foot orthosis provided in this repository, or design another simple joint orthosis with mass and inertia properties.
- When proceeding to the CMC tool, tick the "Actuator constraints" checkbox and provide your device controller .sto file outputed from Step 3.
- The final result files you get from the CMC tool now display the effets of your design. The final step consists in using the "ComputeAFOmetrics" evaluation script to compute metrics based on the 2 different conditions results. Again, one is free to add/modify the metrics computed in the script if needed.
This repository relies on two main folders.
-
The Tools folder house all you need to provide your Matlab environment with the custom code to run the different design tools.
- "CusToM_Excel_process" folder archives all spreadsheets necessary to understand the GRF prediction protocol described in the article.
- "CusToM-master" contains the "Models" folder to insert in your current CusToM install location. The "Mokadim_sim_GRF_prediction" folder stores the resultats of each CusToM simulation to predict subjects GRF in backpack carrying conditions. "OpenSim_Utilities" contains all functions to extract, manipulate and reshape data from OpenSim.
- "markersLocationTF" stores the needed functions to swap from declared markersets in OpenSim to another one suitable with CusToM.
- Finally, "Matlab" folder contains reshaped and augmented GIL data to be used to draw QSC curves (look for "AQS cycle" in "Methods" section of the article for
details), as well as all code to import/extract data from OpenSim .sto files, reshape and augment with subject features, draw and merge curves to get AQS cycles, and compute evaluation metrics. At last, it also includes Simulink projects for all retained subjects and speeds to let designers sketch their assitance strategy and test it.
-
The Exemple folder contains subfolders for each subjects retained from the [Liu et al. 2008] dataset for this framework exemple. Each folder follows the storage system below : Subject(GIL 6/8/11) > Speed(free/slow) > Load Condition(noLoad/Backpack/Backpack+Orthosis) > OpenSim tools results and models. A scaled version (raw and RRA adjusted) of the default Gait10DOF18muscl or modified version for the purpose of this experiment is also present in each subfolders.
Along with the models, each folder stores the outputs of OpenSim analysis tools.
- Analyze : Raw Kinematics and Markers coords in the world frame. Mostly to be used by the transform functions in the Tools folder, to turn kinematics from one markerset to an other for CusToM usage.
- CMC : setup files including tracking tasks and reserve actuators, plus results folder.
- datafiles : Data (Kinematics and GRF) provided by the dataset. Eventually may contain the CusToM-generated GRF for backpack conditions and the Simulink-generated orthosis control signal.
- IK : Inverse kinematics setup file and results computed with the provided data
- RRA : Reduce Residual Algorythm setup file and results.
The dataset used to demonstrate this framework is drawn from a project available on SimTK forum and presented in [Liu et al. 2008]2 Only a part of this dataset was used to run a demonstration of the framework.
The results presented in the article are stored in the "Exemple" section of this repository. The folder structure is explained above.
Citations :