Team Members

  • Yu Hsiao
  • Aaron Wayne



There has been an ongoing debate as to which running form is optimal for endurance in long-distance running. Numerous personal accounts for improvement in performance have been reported by everyday athletes to world champion triathletes describing a number of modified components to their running form that have helped make them run faster and more efficiently, but these factors have not been well-documented scientifically. Though this project is only a tiny step toward analyzing differences in running forms with preliminary data, it provides a platform for future biomechanists to examine the key differences in running forms of different individuals.

This is an extremely difficult question to answer since there is almost no way to experimentally test the differences between running forms. This is because it is impossible to have adequate controls. Every runner is different in height, weight, fitness level, muscular strength, etc. Every runner also runs differently. It’s not possible for one runner to replicate many different styles in an experiment, as every runner has his/her own natural style. However, if different running motion kinematics can be replicated with one runner model in simulation, the differences may be observed.

Using the OpenSim software, simulations can be performed using the same runner and replicating different running styles. Computed Muscle Control also allows muscle activations to be calculated and compared. An additional feature is the addition of a custom muscle model that fatigues to mimic an endurance running event. Since there is no OpenSim model with fatigable muscles, there are many questions that can be investigated. Differences in running form may be analyzed without considering muscle fatigue, but including muscle fatigue may provide more insight.

Research Questions

The objective of the project is to utilize the example fatigable muscle provided in Confluence and embed it in the runner model used for Hamner’s studies [1].

The questions we set out to answer relate to the difference in fatigue between a forefoot striker and a midfoot striker:

  1. Does one running style fatigue faster than the other?
  2. If so, does one require more activation over time to compensate for the fatigued muscle?
  3. Which runner is more efficient?

Modeling and Methods

We used the full-body model described in Hamner's paper [1].

The model provided in the download package for Hamner's Muscle-actuated Simulation of Human Running includes 92 actuators, 76 of which are muscle–tendon actuators based on the Thelen 2003 Muscle Model. The model has 29 degrees of freedom, with each lower extremity having 5 degrees of freedom. Since the model uses Thelen2003 muscles, there was some challenge in converting each muscle into a fatigable muscle.

The original muscle model was to be replaced with the example customized muscle model provided in the Developer's Guide on Confluence. Each of the 76 muscles was changed to a Millard2012Equilibrium-based fatigable muscle model by using a C++ script to modify the model's XML file.

The fatigable muscle model is based on the work of Liu et al. [2]. The muscle has three states: rested, active and fatigued, and the dynamics are governed by the following differential equations:


\dot{M}_A(t) &= BM_U - FM_A + RM_F\\

\dot{M}_U(t) &= M_0 - M_A(t) - M_F(t)\\

\dot{M}_F(t) &= FM_A - RM_F\\

\text{where} \quad M_A &\ \text{is the percentage of activated muscles}\\

M_U &\ \text{is the percentage of unactivated muscles}\\

M_F &\ \text{is the percentage of fatigued muscles}\\

B &\ \text{is the brain command signal}\\

F &\ \text{is the fatigue factor}\\

R &\ \text{is the recovery factor}


The motion files used for the two runners are a forefoot striker and a midfoot striker.

The running motion was first simulated with only one gait cycle and no fatigable muscles. This simulation has been adequate in the past since there should be a minimal amount of fatigue generated in a single gait cycle. When long-duration running is involved, however, new simulations are required, which introduces a couple of challenges. First, the fatigable muscle must be added to the Hamner running model by changing the model’s build file and adding a plugin. Second, since each gait cycle involves a simulation requiring Computed Muscle Control (CMC), the computation time is quite large. For a runner to experience fatigue, he or she may have to run at least 10 minutes to experience any degradation in muscle forces. Therefore, simulations may have to simulate a 10-minute run segment. Assuming a runner runs at a cadence of a modest 85 strides per minute, this requires 850 single-gait-cycle CMCs to be executed. This computational cost is too high and presents a great challenge in finding the right simplification of the model to resemble the real-world situation as accurately as possible. To simplify the simulation, the fatigue factor was greatly increased and the recovery factor was reduced to produce a large amount of fatigue in a short duration.

Results and Conclusions

Two kinematic running motion files were obtained, one from a forefoot striker and one from a midfoot striker, both of whom had very good posture and running form. CMC was run on each model with a simulation duration of 1 s; each simulation took about 4 hours of computation time. Muscle activations were then examined. We first studied differences between a non-fatigable runner and a fatigable runner.



Comparing gluteus maximus activations, the exponential decay in activation can be observed as the result of the dynamics of the fatigable muscle. Upon further inspection, however, the magnitudes of the two figures don't quite match up. This is due to the fact that the original model uses Thelen2003 muscles while the fatigable muscle is based on the Millard2012Equilibrium muscle model.

Next, the fatigable activations are compared between the forefoot and midfoot strikers.



The forefoot striker engages the calf muscles (mainly the soleus) much more, since the runner bounces on his toes. Note that the activation observed in the forefoot striker is more than three times greater than that observed in the midfoot striker. The duration of the engagement is also substantially longer for the forefoot striker. In the forefoot striker, the calf is engaged during the landing for shock absorption and during toe-off for propulsion. In contrast, the midfoot striker primarily engages the calf muscles during toe-off.



Now looking at the quadricep muscle activations, we find that the magnitudes are very similar. But, again, the forefoot striker seems to activate the quads much more than the midfoot striker. Since the forefoot striker lands in front of his hip, his foot is in contact with the ground much longer. Therefore, the quads are needed for more time to stabilize the body during landing and toe-off.



Examining the gluteus maximus muscle group, we find that the forefoot striker activates these muscles much more than the midfoot striker. The duration of each muscle activation is longer, meaning the runner is working harder to generate the joint torques required to sustain this running motion at this speed.



For the hamstring muscle group, the midfoot striker seems to activate slightly more than the forefoot striker. This can be attributed to the fact that, since the midfoot striker lands under his hip, the quads are engaged less for supporting his body weight and the hamstrings are used more for toe-off to create propulsion. Since activations don't differ that much, it could be seen that hamstring muscles are effective for creating propulsion when the stride is done correctly.



Looking at core muscle activations, the midfoot striker again required less activation, shorter in duration and slightly lower in magnitude. This implies that the midfoot striker's gait form is efficient, as the core muscles don't have to work as hard to hold the body upright and stabilize it during a full stride cycle.

It can be clearly seen from major muscle groups (the gluteus maximus, quadriceps, calves, and core muscles) that the forefoot striker requires far more activation than the midfoot striker. It can be concluded that, since the midfoot striker requires less muscle activation, the muscles are producing less force, meaning that the runner uses less energy. Since both runners are running at the same speed, it can be concluded that the midfoot striker is the better runner and the forefoot striker will tire faster with the higher activations. This is significant because the simulation duration was only 1 s. Although the fatigue factor was increased (while maintaining normal fatigue rates), the accumulation of extra use of activation seen in the forefoot striker would accumulate tremendously.

Since the simulation was only performed for 1 s, it was not possible to determine which runner fatigued faster. This would require longer simulations.

Limitations and Future Work

There are obvious limitations to our model that can be improved. Reducing the number of actual gait cycles involved in an endurance running event to about one stride (i.e., using one-second simulations) and increasing the fatigue factor are major limitations in this model. It takes at least four hours to run each simulation, so simulating a longer duration would be too time consuming. Thus, the slow build-up of fatigue through moderate running intensity cannot be studied. The fatigue pattern between an aerobic and an anaerobic intensity cannot be simulated in this model either, as the fatigue rate cannot be changed.

With the example fatigable muscle model provided, the model doesn't account for the composition of individual muscles. An individual may have more or fewer slow-twitch muscle fibers compared to fast-twitch fibers. Varying compositions of slow- and fast-twitch fibers may produce different fatigue rates.

This project also looked at only two running styles. If more styles were simulated, more insight might be gained. It is possible that, for a generic human model, there's a convergence in the most efficient running form that each runner should aim for.

Though we aimed to reproduce running motions objectively with motion files, the uniqueness of each segment of each individual's body was neglected and the motions were replicated with a generic model. Varying body segment length proportions, such as length of the tibia compared to the length of the femur, may produce different joint torques given the same muscle forces and segment masses. This could substantially affect our results.

Through our example, we hoped to provide an easy way for developers to implement a fatigable muscle into their own simulations, following our example using a running model. We are also hopeful that there can be a friendlier path to build a plugin for a fatigable muscle into the OpenSim GUI.

Please visit to download our software.


  1. Hamner, S., Seth, A., Delp, S. (2010) Muscle contribution to propulsion and support during running. Journal of Biomechanics 43, 2709–2716.
  2. Liu, J.Z., Brown, R.W., Yue, G.H. (2002) A dynamical model of muscle activation, fatigue, and recovery. Biophysical Journal 82(5), 2344–2359.
  3. Dreyer, D., Dreyer, K. Chi Running: A Revolutionary Approach to Effortless, Injury-Free Running. New York: Simon & Schuster, 2009.