all right so again welcome to the open sim webinar my name is Jennifer Hicks I’m the open sim R&D manager and I’ll be serving as the moderator for the webinar today I’m pleased to welcome today’s presenters Massimo Sartorius joining us from the University Medical Center göttingen in Germany and Claudio Ziva lotto is joining us from Griffith University in Australia we’re coming from all over the world today so first let me tell you a little bit about open sim it’s a freely available software application for visualizing musculoskeletal structures and simulating movements of humans and animals application includes tools for general purpose inverse dynamics optimization to estimate muscle and joint forces methods to create simulations for motion capture and then tools to analyze and visualize the result of your simulations so the first goal of our webinar series is to showcase cutting-edge research that is being performed with the open sim software platforms open sim is also a growing in geographically diverse community of users so the second goal of our webinar series is to provide an easy platform for the open sim community to communicate and collaborate before we get started I have a few quick reminders about the format of the webinar questions will be addressed at the end of the presentation using the Q&A panel so that’s the first reminder and then the second if you need additional technical help you can consult the guide on our website so now I’d like to introduce Massimo and Claudio first last note Sartori he is a postdoctoral research scientist at the department of neuro rehabilitation engineering at the university medical center göttingen in germany this Reacher’s research focuses on the development of methods that bridge the gap between neural and functional understanding of human movement and then he’s working to translate these developments to advanced neuro rehabilitation technologies we’re also excited to have Massimo visit us here at Stanford for the summer as a visiting scholar in the National Center for simulation and rehabilitation research in addition to Massimo Claudio people ATO is joining us from Griffith University claudio received his bachelor and master’s degree in mechatronics engineering at the University of Padua working with Massimo there as well as Monaco reggiani now he is completing his PhD at the Centre for musculoskeletal at Griffith University under the supervision of Professor David Lloyd his work is focusing on software optimization and real-time application of neuromusculoskeletal models so we’re excited to have them both here with us today to give this webinar on muscle excitation driven musculoskeletal modeling and future applications to neuro rehabilitation technologies so with that take it away Massimo and Claudio alright so thanks nice introduction and hello everyone it’s it’s a real pleasure for me to have the possibility to give an open single webinar today so with this presentation I’d like to talk about part of my research on sort of Oscars creative Ottoman and it is I also would like to give an insight on how some of this application that will be presented and be applied to develop effective solutions for rehabilitation technologies also towards the end of the presentation Claudia who’s currently in Australia will give our demonstrations on how some of the methodologies that I will be presenting in this talk can currently be interfaced with will be the same so the problem of understanding of a movement is a fundamental question that spans the field of neurophysiology and advantage annex and a main limitation here is the current inability of bridging between the in Europe is a logic events happening the humans such as at firing of spiral motor neuron to the spinal cord and the associated muscle skeletal funds the upper or lower extremity which muscles are activated to activate multiple degrees of freedom simultaneously so the ability of bridge between the school world is is important to truly understand the human movement and the beautiful understanding the mechanisms underlying human locomotion is in order to develop effective solutions for individuals who are affected by inter

muscular impairments so a potential a solution for bridging between these two world is given by euro musculoskeletal modeling and this in fact allows converting estimates of neural excitations into predictions of the associated musculoskeletal function and that’s what I’d like to talk about today in this in this presentation so the very long term in the title muscle excitation driven musculoskeletal modeling simply means that we will be modeling the dynamics of the muscular and the skeletal system is controlled by the variable system so that here here is to collect experimental data reflecting a dynamic of of the neuro drive muscles and use these informations to perform a master driven simulations to understand how muscle contract generate force and contribute to prove contribute to generate and final movement and in this context it is possible to collect well what most cite could you increase the volume on your microphone I’m having a hard time hearing you but are you can you move a little bit closer to the microphone that’s a little bit better look that’s a little bit better yeah okay so in this context we can actually record experimental data reflecting the dynamics of the new exact amount of the different levels and in the you must have pathway so at the beginning of my presentation I will personally talk about how to use experimental recordings of energy signals to characterize excitations and use this as an input muscle scraping models I will then talk about how to how it is currently possible to reduce the complexity in of the energy signals by exploiting the theory between muscle villages and this allows us gaining an insight how the central plasma generating which is linked encoding the sponsor futures act towards the activations of multiple muscles to bring a multipass finally I will show how we can use high-density electromyography to gain an insight in how individual motor units into the muscles are activated and it is a great opportunity to have a direct estimate of the activity of motor units more team uses on a pathway and it is very help better understand the mechanisms must replicate in the production so in order for this to happen it is very important to develop multi-level models did account the new physiological level and the musculoskeletal level and that’s important because although we have little knowledge about what happens at one side in what happens at the other side we still do not have relevant information or social advanced one level to advanced at the older level so in this presentation I will introduce our expectation driven musculus written model I will show how we can use estimate of mural excitation is recorded at different levels in the near amount of pathway and use this to predict the resulting function from the musculoskeletal system I wouldn’t show how this can be this can be used to implement effective swings in and then cloudy at the end to have our model can be so I like to start off the presentation by explain how we can use experimental mg data as an input type in C data are probably the most viable way to record the ability of a large number of muscles during a highly dynamic task and it is complex our model is able to use to produce all the transformation to

take place in the onset of of mgs until the production force into multiple muscles and joints into multiple degrees of freedom its benefit of using injected modeling is the possibility of performing massive giving simulations without making assumptions on the way muscles activate and share a load across across the and that’s because experimental EMGs which actually reflect the actual activations of the muscles during the recorded movement are used as direct input drive in the model and it is very important to characterize a number of mechanisms in the muscles that could be difficult to be characterized otherwise including muscle preoccupations and muscle contraction so as we can see the model requires input data reflecting the human movement and in our studies in Stata we’re recording using a gait laboratory so we actually use surface electrodes which we place on the measure muscles in the lower extremity and when we use will be placed on the subject retro reflective markers and we recorded the three-dimensional positions of this marker using multiple cameras surrounding the subjects movement and then the subject was able to walk on an inground force plate that we use to record the food ground reaction forces generated through the movement and all these experimental data were then use here with OpenSim to produce a number of input data that we needed to calibrate and validate the model and let explain this later so with the next slide I would like to spend a little bit more of time to explain what happens in this step and then I will talk about how the model works move to the next slide so I’d like to stress in fact that in Jeju the model that we’ll be presenting in the next few slide was developed starting from the previous conclusions of many people and in many institutions and this Waialae to to thank if not all of them at least some of them that didn’t overhear the characterizations of the human movement from experimental data was composed of four steps in our in our methodology so in the first step we use market trajectories recording to bring a functional task to locate the joint centers of rotations of the hip knee and the ankle and we also locate the helical Technic second axis at the knee and is using a solving a numerical optimization problem in the second step we take a generic OpenSim musculoskeletal model and scale it up or down to match the interpolation of a subject in the third step we use marker trajectories recording during dynamic tasks and we solve an inverse kinematics problem from which we get estimates of three dimensional joint angles and then we we use this estimate together with ground grappling forces to solve an inversion any problem that we can use to actually gain estimates on on joint moments and we do this using dynamic simulations of motions I actually hope you can see this in writing Qwest smoothly despite the limitations in broadcasting of of WebEx so at the end of the simulations we have experienced experimental two angles that you need later on in the model and three moments for calibration validation so at the end of this step we have all the input data that we need to run the model so we have two angles extracted from market trajectories from an inverse kinematics in legends in open sim and we have EMG signals record experimentally and that’s what the models look like so the model is composed of four main blocks as we can see in the first block even signals are converted into muscle muscle activation so in this block will be basically account for the electromechanical delay which is basically a the time difference between the onset of the electrical activity in the muscle and the actual time where must physically contract and produce force and in this log we also account for the

non-linearity that exists between muscle expectations and mass of course in the second block we use estimates up to an angle to compute the kinematics of muscles including including muscle tangle length and human arms in the third block we combine together muscle activation to muscle panel lengths and we generate estimates of the food produced by the muscle and once again 18:8 the force produced by a muscle in a specific instant of time we can then project this force onto multiple domains onto multiple degrees of freedoms and do this by combining the force with the respective Norman arms so in this way we can understand how a single produce by a single muscle can contribute to the simultaneous activation of multiple degrees of freedom in the human activities the motor can which is estimated muscle force and twelve moments using twin angles and EMG data if we look around the model though the model has to be calculated and calibration is necessary to determine a number of parameters that characterize how muscle activates and how Matthew contract now these parameters vary linearly across subjects and they are very difficult to be derived from the new church or experimentally used to use an optimization approach and there is actually start of this optimization process giving the model generic set of parameters that describe an average subject and the model we use these parameters together with input data taken from of congressionals and we’ll produce estimates of muscle force enjoy moments and in this step we now have a third input which are the experimental joint moments and the day is to continuously adjust the model parameters until we can minimize the mismatch between predicted and experimental joint moments and which is over a variety of motor tasks including forward walking running crossover and sidestepping at the end of this step we have an optimal set of parameters that allow the model can run up on loop so now the model is able to produce dynamically confidence in estimates muscle force enjoy moments typically as a function for toy angles and modularity and during the validation step we round them although using novel motor coils that were not used during the calibration and this trial reflect a different way of contraction strategies and the idea was to see if if the model was robust enough it could be able to predict dynamically consistent joint moments about multiple degrees of freedom during novel trials and to account for different ways muscle activate and this would make the model of fully predictive system mirror some experimental results that you how the calculated model running in open loop was able to predict the joint moments produced about 6 degrees of freedom in the lower extremity so this is the hip second extension keep attacking abduction keep internal external rotation knee flexion extension angle planted or zigzag human ankles of dollar flexion we can and this was done during cross sidestepping work walking and running as we can see the model was able to use him G data and predict estimates of 12 months that matched the reference and the interesting point here is that the calibrated model is flexible and to a certain extent is also generic because it can be operated on a variety of tasks they are bio mechanically different with each other’s and angulation muscle contraction strategies so this flexibility and the calibrated model is quite important in in the context of rehabilitation technologies with the next so I would like to talk a little bit about about this the key point is that the energy dependent predictions of the muscle force enjoy moments can actually be used to implement proportional control in in a skeleton and in this example here we can see how

a very simple engineering model of the knee joint with it to use even signals taken from the extensor and reflex or knee muscles and convert the experimenter images into estimates of muscle force in a joint legend exact moment and we can see here on this graph the green bar represent the flexion extension moment predicted from from the EMG signal so this is actually the strength in subjects muscle and the blue curve I hope you can see it is the additional work that the exoskeleton is supplying to the subject and that’s the terminus functions of the work predictor from the subject so the interesting point here is that empirical modeling allows understanding the strength of an individual and this is important to set the exoskeleton support in order to provide additional to work as much as it’s needed so this allows keeping the subject active the rehabilitation treatment and to engage so using a method as an inputs for our model we can predict output variables that can be useful and it can be effective in implementations of control system in exoskeletons in this context we need up to 16 electrodes to cover the main muscles in the lower extremity and that the many degrees of freedom now having seen sensors be a limitations in in the context of real applications daily life details and next slide I’d like to talk about another methodologies to develop and this methodologies actually allows removing this is high number of sensors and with this methodologies we’re actually driving the model instead of using experimental estimates of data we’re driving the model using a set of Gaussian curves which are completely parameterised as functions of of the Caged cycle and the thing is to use this set of Gaussian curves to define very simple strategies of muscle recruitment that we can use to simulate muscle activations throughout locomotion and I will explain later this market strategies are actually reflections of the central part in generating networks which are believed to be encoded in the spinal cord so going back to the scheme that summarize the structure of this presentation we now are working at this level so we’re going to use muscle synergies reflecting upon secretaries as an input to the model must allow reducing the dimensionality of Mg signals and the basic idea is that the same modularity then that exists in the spinal cord can also be observed at the muscular level – so this means that the recruitment of a large number of muscles and can be explained by a small number of excitation primitives and the key point is that this modularity can also be observed directly from EMG signals so in this scenario the EMG signals that we are able to require using surface detection biography are the result of the activations of individual muscles filtered by by the skin tissues and affected by by crosstalk so according to the muscle synergy theory and the activations of individual muscles are not independent with each other but they are actually explained by the activations of a smaller set of attention primitives which are combined together according to some rules for the motor modules so the linear combinations of these small number of excitation primitives allows reconstructing the large repertoire of activations that we can observe in G signals so in this according to this idea in this study we’re not using in Jessica’s anymore as an input for the model but we are using it is exiting pivoting now due to time constraints I’m not gonna explain how we can derive these primitives from EMG data but I’ll be more than happy to further discuss these in the presentation if any of you

is interested so we now have this low dimensional set of excitation primitives and this determine any need recruitment of muscle groups in our model and the use of the terminology of even encode colleagues the set of dodging cars can be seen as an impulsive controller because the old set of 1 excitation privat is basically determined the recruitment of a group of synergistic muscles so as the gait cycle progresses these exiting primitives will will spike and will generate an impulsive and sequential include recruitment or group of muscles so the key here is to refine this in recruitment better fit the the muscle construction strategies observed across a variety of tasks of motor tasks and which it is by processing this initial excitations using a second order recursive filter that we use to simulate the muscle twitches in response to the initial engine excitation and then we use an energy transfer box to a comfort and linearity between excitation muscle force and the a you receive rewards to what I previously explained here we adjust a number of muscle specific parameters until we can minimize the the joint moment prediction error across our mantra tasks and at the end of this process we have a final mapping between the 5×11 primitives and the 34 muscle can use it in the model and I like to stress the fact that this mapping which is with the rise from this calibration basically represents a best fit oh the recommended strategies that were observed across this task and that’s because the calibration process generated a single set of parameters that describes on average I did an average of the forecast but the better explained is I will show the example of of the peroneus muscles so here we can see the green curve is the the muscle activation predicted from the impulsive controller and we can see that the activations of the muscle is the same across the for motor tasks so that’s all the same and we can see that it only feeds on average the activations produced by in G’s and our results show that although it is impossible is very simple and it is nothing in the sense that it’s always the same across the school motor task when the combined is imported controller with accurate estimates of masu kinematics then we are able to produce dynamically consistent estimate of joint moments with comparable accuracy using jisuk other than input here are some experimental results showing this and who as we can see here the knowledge driven by data technical primitives was able to predict when moments about six degrees of freedoms we cross over sidestepping for cane and running and we can see that the prediction accuracy was comparable to that achieved by using a machine as an input the interesting point here is that if we focus on one degree of freedom when we compare it across tasks we can see that the joint function predicted by the model was quite different across the full motor tasks and this was predicted by having the same impulsive controller as an input across and for motor tasks and it was achieved not just during 1 degrees of freedom and across all the degrees of freedom so this suggests that dynamically different motor tasks may actually share new master control strategies which are similar and are impulsive in nature so after now I talked how we can use estimates that indirectly reflect the neural Drive so we used EMG signals and muscle synergies which are an indirect representations of the newer driver muscles and with excuse like I like to talk about how we can use even range motor units as a bit forward and cooler the inputs right our model and that’s interesting because motive units are directly related to the activity of motor neurons in the muscles but the day here is to use a

bi-dimensional electrodes and to record activity of the muscle so this is called high density election biography and this allows us to record the activity of the muscle from different locations simultaneously so this gives us an example of the signal and a good analogy here is that of having a room full of people talking simultaneously and classical election biography which is space and using a single pair of electrodes located in a specific area of the muscle would be equivalent to have a single microphone in the room in a specific point of the room to record the overall and sample of voices of people talking simultaneously whereas tied an election biography it’s like having multiple microphones throughout the room to record voices from different points and it is that once we have two accurate a recording of the example of signals we can decompose it and we can determine the underlying sources the generator is intended this is available a little bit like saying and like trying to differentiate the voice of each individual person from the recording people come to some kind of lip and of course in the terms of muscles each person represents a model unit so applying the composition we can extract a number of motor units in this in this strategy we concede we’re able to extract 50 materials so every row represent a multi unit and every column represents at the temporal event at which that specific monster unit was activated and so this gives us a spy train which which magically reflect activations of neurons in the spinal cord which innervate multiple not view it in a specific moment so by combining together these two spike trains for all the motor units that we were able to move decompose we can extract the cumulative spike train for a specific muscle and we can do this for for an animal crossing a joint we actually did it for five major muscles crossing the joint angle and we can then combine we can then use these cumulative supply trains to drive masters kitten models and we can predict the functions at the joint and our cleaning area results show that it’s possible to use promoted spike trains in a symmetric conditions and this is quite exciting and it’s quite promising so up until now I talked how we can use estimate of muscle excitations try to predict the functions and the mascots cradle system business regarded as a four were dynamic model in this scenario we can use inverse models which reconstruct all the Moscow City the transformations in Reverse so starting from for the final movement we can reconstruct in Reverse all the muscles the transformations and determining the man who controls the generated it and the neural excitations responsible for the generation for the final movement and we did this map using the static optimization that we were able to reconstruct in reverse all the activations until the generations of synthetic engine signals and today he was to combine this two model together so we’ll combine the two for a dynamic model we are in person egg model together we’re closing the loop and we now have what I call a hybrid model and this hybrid model allows transitioning from the configuration in which models can be fully driven by experimental in GC goes in a full dynamic way preparation which muscle the most activity can be fully optimized and derived in an impure dynamic way but it’s also allowed for all the graduations in J’s in between and this hybrid modeling approach allows addressing some limitations in service delivery America mutations is the inability of intellectual Yagami of accessing people locate muscles and other limitations is the ability of

properly accounting from for cross talk in muscles and also properly representing and taking bandwidth in neural excitation but depending on how much we trust our experimental images to begin with we can decide to adjust them more or less by moving on this on this line and we do this this process of adjustment the purpose of improving the estimate enjoy moment and the purpose of addressing the attentional selection biography and the point here is to find the minimum level of adjustment that gives us the best improve in joint moment prediction accuracy here we can see in this graph how we were able to minimally adjust little angles for all the muscles in our model so the red cars here represent the experimental changes and the great cars are the linear ad load which were minimally adjusted we consider for many muscles the Johnson mgs are very close the experimental one and some muscles we had a substantial level of adjustment and on top of that we were able to predict to estimate the activity of deeply located muscles like the solos and heliacal which didn’t have experimentally mg’s to begin with and it is that using this minimally adjustable inner envelope we can improve much better predictions of joint moments about six degrees of freedom and the ability of predicting dynamically consistent estimates of all the muscle crossing the hip is really important to study the dynamics of the hip joint so in this case we can have a very nice estimates of enforces producers by all the masks crossing the joint and we can combine this force estimates with finite element modeling and understand how muscles interact with bone and contributes to the remodeling of bone and and this is really important in the fitting internal implants but also for the prediction of the development of muscular disease so with this slide I’m going to start the last part of the presentation that I’m going to spend a few slides to introduce the demo which will explain how our model can currently talk with open system I like to spend effective all the methodologies that I presented in this talk are currently being integrated into a box that has this very long name so I created EMG inform your moscow skeletal model in people and as i showed in the presentation the toolbox is able to reuse estimates of the neural excitations reflecting different level of reflected or drive and different levels a pathway and the pool box allows choosing difference operation modes for for the underlying model so we can decide how we want to implement the mapping between excitation to activation depending on what ex-teammates of master excitation we have as an input we can use different ways to estimate the model kinematics from from joint angles we have the possibility of using a different angle models we can also decide how many degrees of freedom we and how many muscles we want to run our simulation with we can then decide to operate the model as a word then adding models of the driven by excitations or as an investor Namek model so we can use static optimization or we can run the model as as an algorithm we can then determine and decide what parameters we want to calibrate in the model and what optimizer we want to use and finally can decide if you want to execute the calculated model offline or in real time for applications of control of assistive devices and all these variables can be dealt by XML I don’t like it’s done in consume at the moment so as I said before the knowledge can currently be used with an organism and in the next to like card we explained how we were able to achieve this level of interface between our two analyzing so I now would

like passable over to Italian Thank You ma Ciro hello everyone I hope my microphone works of course fine so as maximally explained as cinemas is a powerful tool for the simulation of muscle dynamic as we can see in this slide it is possible to choose different options and flavors – ants in a mess however for the integration of cinema sequencing we selected a specific combination of these options like the use of Ian’s Jesus input jelastic implementation of d-10 the dynamic and the use of multiple degrees of freedom finally we selected the forward dynamics mode which means that cinemas will work in open-loop full predictive mode what we want to achieve is an exchange of information between open scene and cinemas open scene computes muscle tendon length and moment arms which are used with experimentally recorded EMGs as input for cinemas consequently cinemas computes the muscle dynamic according to the options previously defined and returns to open seeing the muscle activations and muscle forces state the goal is to allow the user to perform EMG informed simulation within the open scene GUI and receive visual feedback of the results to achieve a complete integration of cinemas in open seam we firstly wrapped cinemas with the open scene api’s starting from emotional file with encode data the API is performed the muscle analysis and feed cinemas with muscle tan lines and moment dance secondly we created a personal ID from 10 to allow the user to select the dictionary input required by cinemas as experimental EMGs finally we used the new weapons in scripting feature introduced with the 3.0 version to create a communication layer between opens in Japan seen GUI and the other software components therefore the user is able to easily interact with cinemas and has access to all the result directly from the bingo now I’m going to show the result of this integration with a brief live demonstration so here is open scene and first of all we are to open an amazing model this model as you know defines the geometry of the muscles and consequently will influence the result of the simulation every valid open C model can be used of course second on the script drop-down menu we select our script runs in MS and so our custom wings of appears in this window we are able to see which is the currently active model in open sim and now we can select the open theme motion motion file that contains jingle all the angle of our motion secondly we select the EMG file that contains the experimentally EMG data relative to the selected trial and laughs which is the XML file that contains the subject specific muscle parameters which have been estimated during the calibration process that muscle described before they click on the Run button starts the EMG inform simulation so in this moment the muscle dynamic is being estimated by the api’s and by cinemas here it is the simulation is just finished and we can see the results of the object evasion of the muscle so we can clearly clearly see which muscles are being activated and this is from cinemas from our simulation so furthermore it is possible to view the result of this computation directly from the open same standard plot tool yeah so it’s possible to to plot

activations or forces this case activations here it is these are deactivation that we competed just before then yeah we can see the the line moving on the plot so yes this is oh and it’s possible to you to plot the forces also the forces that the results of our simulations of the busses it is so this finishes the the part will open same let’s go back to the slide and so as we have seen this integration between cinemas and often seen makes available mg form simulation directly from the open seam GUI and before before letting Massimo to conclude the presentation I would like to thank Monica and Johnny she is one of the main developers of cinemas along with Matthew and I and thanks to her we have such a modular powerful talk the software at cinemas so now that muscle to finish and to give a brief summary of what we have seen today all right demonstration I think you forgot to say that our tool box will be made available on cindy kaye and to begin with it will be made available as internal executable so with this slide I’d like to summarize what has been said so far so our applications based on collecting experimental data reflecting directly or indirectly then you attracted muscles and uses a CMS of muscle expectations within our model and to be able to predict the resulting functions in the damascus cradle system this gives us the possibility of using muscle excitation to understand how models interact with forms but also to implement computer control in assistive devices including exoskeletons or or prosthesis we can exploit the modularity in muscles to simplify the complexity of EMG data added and it allows us to implement proportional control devices even when Mg data may not be entirely available from the patient’s Maps finally we can use high-density electromyography to gain an insight into how individual motor units activate within a muscle that is important to better understand how the new logic world is connected to the master script world this can also have important implications for the development of my electric control in prosthesis and poverty so with this light I’d like to conclude our presentations and I’d like to thank you for your attention all right Thank You Massimo and Claudio for a great talk we can now go ahead and get started with the Q&A session so here’s a little diagram to show you how to ask questions so you’ll use the Q&A panel so all the questions will be text-based you want to find the Q&A panel and your WebEx controls and then type in your question and make sure you choose to ask all panelists so with that we’ll go ahead and get started we already have a couple questions that have come in the first is the practical one from Jing psy and his question is how much time does it take for the model calibration process to complete and I’ll actually add on a second question related to that so how long does it take as well if you you know if you have you calibrate a model to a particular person if they come back a week later a month later do you have to recalibrate I assume you do since it’s that EMG placement might have

moved but you can you guys can comment on that sometimes were great questions the calibration process it’s quite quick and and that’s also thanks to cloud in Monica Johnny we really optimize the code so in the depends on how we operated model so the calibration time varies as a function of the tandem model use if we use a system model the calibration on one subject using six to eight trials the calibration it runs in two to three minutes and using an elastic tangent it takes longer and I actually let cloud you tell how much you would take with an elastic oh well doesn’t take so much longer it’s about 1215 minutes maximum so with the same setting as you as you described so yes and so calibration is quite fast and the execution of the model is looking running real time runs within the massive electromechanical delay so we can have an SMS of muscle force enjoyment every 20 milliseconds and with regards to the addition of of Jann it has been proved that so the parameters that are calibrating the model can be divided into two groups so R&R of parameters that determine the EMG activation transfer function and there are a number of parameters that determining the knowledge of the muscles so this is kind of leg length as optimal father length and whereas the mass is related to the EMG to activation transformations are we have computering coefficients which determine how the internal filter of the model respond one impulsive input and we then have a parameter which determines the non-linearity between excitation force and what we were worried was proved by the work of David Lloyd and Tom Buchanan and this year and colleagues is that it is possible to calibrate once all the unequivocal parameters that in that meaning tendo’s LED lamp an optimal path length because those are related to their geometry to get at me of a subject and in a now in on an adult subject responders don’t change so this parameters need to be calibrated once on the other hand the parameters that are associated to the engine activation transformations they are quite sensitive the placement of the images and also a number of physiological conditions like low pressure in it is own disappearance so these ones have to be correctly calibrated to every decision ok great thanks to a very clear answer to those to those questions so we have a couple more questions related to calibration Massimo’s can I remind you to really speak up and speak into the microphone just because there’s there’s a little bit of a difference between your volume and Claudio’s volume so ok we don’t know we don’t want to blow people’s ears out ok so the next question is from Stefan Lambrecht his question is what is the influence of the various calibration tasks that you use a running sidestepping crossover on the predictive value in other words how does the accuracy improve by including more than one task in the calibration right so basically the idea of adding tasks which are by mechanically different with each other it’s it’s not so we don’t do this because the purpose of improving the accuracy of the calibration we do this for the purpose of accounting for a large a larger spectrum of muscle recruitment strategies the calibration would work the best if we use just one file for eBridge this would give us a perfect fit I think this will give us a parameter set that fit that specific file

perfectly but that would happen is that when we run them or they’re not pollute the model will not be predictive because the mother would be the model we take that lately – just one one specific trial so in order to improve the predict the predictive ability of the model we include a larger variety of tasks that allows us to represent the full spectrum of mathematical things but this in principle decreases the overall this actually decreases the this increases the the RMS error between predicted and experimental images but on the other hand it also increases the predictive ability of them also we can run the model on many of its failures and the model will will be that we need to stay dynamic and dynamically consistent okay great so is the calibration process is that going to be part of the toolbox that you’re going to share on cindy kaye yes yes idea is to also provide calibration as as Google probably cloud you can say a few more words to say what we are planning to release in the future yes so the jdi’s to release is these two books even though we want to be sure that everything runs smooth for the user and there are no usual problems or anything like that but we are planning to release the calibration as well because otherwise the model can’t work properly without a calibration process and we I’m going to implement the calibration in a similar way as we done with the with the execution part of the model itself so what we have seen there will be something similar and easy to use as the execution of cinema’s the estimation of the mass of roses so it will be similar thing still with the scraping and yes will be available in this I think we need some mom’s still to be sure that everything is just perfect so that’s the shower plan great but I bet answered another question that had come in when when we were planning their release what we what we want to do is before we release the the soft we want to make sure that the software is fog free so this is going to take we’re going to need a couple of extra weeks probably a few months but in the end we will have first release of the box cindy kaye and this will feature ing driven model running great sounds great ok so let’s let’s move on to another question let’s see where did it go ok so this is a question from schemas two mounted through apologies by I mispronounced just great talk thank you can you comment on considering mechanical behavior of the muscle introduced in the GFCI NMS toolbox in other words to predict the forces through an electromechanical model using a constitutive model plus activation inputs that you referred to I’m not sure if I understood the question the question is related to how the process of excitations of muscle and force production works yes I think the question is whether you considered using a constitutive model of that process right so what we do at the at this time we have a blog that models the transformation between X between excitations and activation and as I explained this is important to account for the nonlinearities in masks which which actually may vary across muscles

and then of course our our master model is is a hill type muscle model so it says it’s a it’s a it’s a standard Hill that massaman which actually was modified by the many people who contributed and hi actually acknowledge them at the beginning of my thoughts of these features a nice elastic tandem dynamics a nice way to predict the optimal fiber length and how this change into the functions of muscle excitations but for the moment we’re planning to keep this model has has has main muscle implementations in our crew box but we’re open to extend different okay I know we have a couple questions about the parameters of your model first from Wally’s Hasani he’s wondering how you know that the results you say that you get say for muscle optimal lengths are a good value since you’re just optimizing to two fitted data and related to that Jonathan Lewis asks how much do the parameters typically very when you use different calibration runs for different people yes so it’s a very nice question so producer questions so the key point of the calibration is that it is a constrained optimization process this means that parameters are allowed to vary the physiological boundaries so we don’t allow para parameters to very indefinite Li so for example a tennis leg length an optical fiber length we we can only adjust them within a 2.5 percent of the initial estimate and this initial estimate of x leg length an optical fiber length is derived using a methodology which was published by windy and colleague sponsor of biomechanics forum in Rome 2008 this is basically I will be like the inner schedule OpenSim in this case we are actually able to scale the known in early the optimum fiber length and at the end tendo’s leg length to achieve a nice starting point and then these parameters can be further adjusted in the calculations and they only are allowed to vary between a 2.5 percent of their initial value and same story for the EMG to activation parameters so we have two filtering conditions and parameters they are also constrained they vary between minus 1 and 0 and it is to basically keep the filter stable and then we have the no linear shape factor which varies between minus 3 and 0 and this was published by Hannon and colleagues it is constrained and what he showed is that there is consistency in the calibrations of the anatomical parameters vary and where the calibration of the energy to Tyrian parameter this is less is way less consistent so by using data from different data set we would have substantially different results in those parameters okay great right thank you for for clarifying that I think we are about out of time so I’m going to go ahead and close the Q&A session thanks for all of your all the great questions and your clear answers Massimo and Claudio can you go ahead and move me to the next slide the next one yeah okay so as we definitely want to acknowledge the funding sources for the open sim projects open ximena’s webinar series are supported by several grants from the NIH and the EU including an NIH grant that funds our National Center for simulation and rehabilitation research we’re also supported by the dark warrior web effort I want to again thank you all for a great discussion you can find information about the Center upcoming events future webinars and other resources from the open sim community on our website and then also please fill out the survey that will appear in a pop-up window at the conclusion of the webinar this will help us improve the

webinars and choose future upcoming topics so thank you again for participating in our webinar and we hope you’ll continue to stay involved with OpenSim thank you you you