ASMI Limitations

teddyrooseveltIn June 2004 I was working as a Senior Research Scientist at Callaway Golf.  I was responsible for leading research program investigating the Product Player Matching initiative put forth by senior management.  The projects that I was in charge of included Player Profiling, Digital Human Modeling, and Club Fitting.  A central component to these 3 projects was an inertial measurement unit (IMU) on-board diagnostic (OBD) shaft-based data acquisition system, that would link all three scientific research projects together.  One of the more difficult aspects of the project was verifying angular velocity measurements between the IMU club used for player profiling and motion capture measurements collected in the Player Performance Bay which were used as inputs to the digital human modeling efforts.  I still remember that day when we verified the dynamic measurements between the two systems were the same.  It was a Eureka! moment as it verified all of the synergistic plans using multiple measurement systems for the different projects moving forward.

Two days later I was sitting in my cubicle and I picked up the June 2004 issue of Wired magazine.  Inside was an article on the American Sports Medicine Institute (ASMI) and the work they were doing on pitching biomechanics entitled The Ultimate Pitching Machine. The first paragraph read, “It’s the fastest human movement ever measured in a laboratory: the violent forward rotation of a baseball pitcher’s shoulder as he whips the ball toward home plate. To the batter, or anyone watching on TV, the motion is merely a blur. But not to biomechanical engineer Glenn Fleisig. “The arm flips forward at the shoulder joint with an angular velocity of 7,200 degrees per second,” says Fleisig. “If a pitcher’s arm kept up that velocity for a full second, it would make 20 full revolutions. It’s just phenomenal.

When I first read that I thought “no way is that value right.”  I was speaking specifically to the magnitude of the angular velocity value, not the fact that the overhead throwing motion is the fastest human movement.  I had just spent over a month testing and re-testing both IMU club measurements and motion capture measurements analyzing angular velocity of the club’s release and closure rates.  We were measuring angular velocity rates of 2,000 – 3,200 degrees/second at the butt end of the shaft of the golf club. This article was saying it was 2-3 times that.  Then I looked at the pictures in the article which were time stamped and basic 2D back of the envelope math calculations showed that it was highly unlikely those values were accurate.  I was fairly certain that I was right but didn’t really have a firm handle on why the magnitude was not accurate. So that was always in the back of my mind, but I was not working on analyzing pitching biomechanics at the time so I just figured someone else could prove it was inaccurate.

Fast forward 10 years later and I still see the same problem.  While there have been some studies that have reported much lower angular velocity rates for the shoulder internal rotational angular velocity, no one has identified the differences.  I have finally started devoting more time to pitching biomechanics studies so it has become a much higher priority to demonstrate why the absolute magnitudes that are published by ASMI for shoulder internal rotational angular velocity rates do not match true humeral angular velocities.  More importantly, we have finally reached a point where more appropriate inertial sensor measurement systems such as SportSemble have been tested and other systems such as the Motus Sleeve may become available to the market very soon.  The big question is are these different systems speaking the same language – i.e. are angular velocity measurements between the different systems providing the exact same quantitative measurements?

Having worked with a prototype inertial sensor system for baseball mechanics, I know first-hand that they can provide similar measurements if proper methodologies are employed.  However, the methodology used by most pitching biomechanics motion capture studies will not allow direct comparison between the two systems.  In this page I will go into detail why specifically the ASMI reported angular velocity rates do not quantitatively match true humeral angular velocities from inertial sensor measurements.

When I first started deeply investigating the methodologies used by ASMI for pitching arm side (PAS) kinematics, and specifically how they measured shoulder internal rotational angular velocity rates, I expected to find that there was a serious flaw in their methodologies.  That was not the case, which was very refreshing given how long I have followed and admired ASMI research studies.  In their early publications, they were very upfront about their measurement system limitations and typically discussed these issues in their published research papers.  I will also say that ASMI published papers are always extremely well written, almost textbook on how a research scientist should present their methodologies and results while also addressing limitations of their study.

So in no way was ASMI trying to hide or deceive what they were measuring and reporting. Rather, they gave very important insight into why they measured what they did and subsequently reported, and leave the interpretation up to the reader as to the effect of the potential limitations on the reported results.  Seems like a very fair scientific compromise given technology limitations in the early 1990s.  It is my personal experience that those limitations are still found in their present day methodologies despite technological advances, and that limits the application of their output data for subsequent use such as in forward dynamics simulations.

I have only had a few brief email correspondences with Dr. Fleisig and he has been very gracious forwarding me copies of published articles that I have requested.  He always offers to send any others if I need them.  He is definitely very well respected and is truly interested in providing the best information for up-to-date research regarding pitching biomechanics and injury prevention.  I want it to be perfectly clear that I am not attacking ASMI methodologies or investigators in this post.  I chose the Teddy Roosevelt picture at the beginning of this page for a reason.  I absolutely commend ASMI  on the work they have done over the past 25 years to further pitching biomechanics research when no one else did.

My intent is not to criticize; rather, I am trying to identify a deficiency in the ASMI measurement methodologies that will limit the ability to quantitatively analyze PAS shoulder and elbow performance and injury mechanisms moving forward.  I work in a very specialized niche in biomechanical modeling and simulation that is completely dependent upon repeatable and reproducible angular velocity and acceleration data. Thus, it is imperative that I assess the validity of the absolute PAS shoulder and elbow kinematic values, as well as subsequent kinetic values which are based directly on these initial measurements.  It is extremely important that this issue is addressed now, as inertial sensor technologies such as the Motus Pitch Sleeve will soon be brought to market, and any research activities utilizing both optical motion capture and inertial sensor systems will require an understanding of these numerical differences.

I have been fortunate to have exposure to the latest and greatest biomechanical evaluation system technologies and techniques that allow me to analyze and point out potentially better alternatives for synergistic quantifiable athletic performance and injury analysis research programs.  I myself have had to make similar choices in measurement technologies and weigh the costs of accuracy vs. budgetary concerns.  Fortunately, I was provided a higher fiscal budget for these technologies working in applied research vs. the non-profit academic research constraints of ASMI.  Given time, the availability of better and cheaper technology now allows even academic research programs to utilize the latest and greatest evaluation systems.  Despite advances in camera technologies that ASMI has implemented in their methodologies, they still use essentially the same marker protocol that was established in 1993.  There is video evidence of additional markers being used in their studies, however no new papers have been published identifying kinematic methodological differences.

There are 3 main areas of limitations in ASMI methodologies that limit the accuracy of their measured and reported PAS kinematics and subsequent kinetics.  These are:

1.) Rigid torso-PAS connectivity: as was described in the ASMI Methodologies section, they use what I consider to be a limited marker set.  However, as was described in Dillman et al. 1993, they initially made that choice because of limitations in camera resolution.  They could not use more markers devoted to the torso and/or upper arm, without taking away markers from the lower body.  As they were interested in providing both clinical and research outputs, they chose the full body marker protocol, realizing there were potential limitations in their marker protocol.

One specific area that suffers is the modeling capabilities of the PAS extremity and connectivity to the torso.  In their methodologies, they use 2 markers at the hips (bilaterally at the greater trochanter) and 2 at the shoulders (bilaterally at the lateral tip of the acromion) for torso calculations.  They use the same acromion marker on the PAS extremity and one other marker at the PAS lateral epicondyle for measuring PAS humeral rotations.  So that provides a very stiff link connecting the humerus to the torso.  SFCxfactorMore importantly, it doesn’t allow for measurement and analysis of any upper torso hyperextension or bilateral scap loading (humeral horizontal abduction in ASMI methodologies) that is clearly exhibited in MLB mechanics.  The measured torso kinematics swallow up any actual hyperextension of the torso.  Also, while scap loading is often symmetrical relative to the spine during the stride, the rapid unwinding of the torso from stride foot contact (SFC) through delivery presents plenty of examples of asymmetrical bilateral scap loading that also gets lost in the torso kinematics based on their measurement protocol.  In Cliff Lee’s picture above on the right it is clearly seen that his PAS shows much higher horizontal abduction than on the non-throwing side. However, because the torso kinematics are measured through the lateral tips of the bilateral acromion, this differentiator in shoulder kinematics is partially lost and attributed to torso kinematics.  Essentially, the ASMI results would show that his torso is closed more than it actually is.

That is not to say that it is an easy situation to measure these differences, as the spine is very flexible and it is difficult to measure scapular kinematics.  However, it can still be accomplished.  Many biomechanics studies will measure and analyze lower torso and upper torso kinematics at a minimum in an attempt to analyze the X-factor stretch between the pelvis/core and upper torso that is so important in rotational dynamics.  Others will even model the spine as three separate segments.  Recently, biomechanics of pitching studies have measured scapular kinematics to analyze the individual contributions of scapulothoracic, extension of the spine, and glenohumeral rotations as opposed to the lumped parameter estimation of shoulder internal rotational angular velocity provided by ASMI.

ASMI has recognized these situations in their published papers.  From Dillman et al. 1993, in the arm cocking phase they write, “Trunk rotation follows the hip but, in highly skilled pitchers, hyperextension of the upper trunk occurs as it is rotated around to face the plate.”  Yet their measurement methodology is incapable of measuring this situation and this potential kinematics differentiator gets lost.  The following picture was taken from Aguinaldo et al. 2009 and provides a good image showing the extreme hyperextension of the upper trunk during the pitching motion that the ASMI measurement methodology misrepresents.  aguinaldo2009Any study that uses this marker methodology to compare different throwing styles should ultimately be analyzed with skepticism, as the underlying gauge R&R is inadequate for measuring any torso differences such as hyperextension of the upper trunk.  That is one of the reasons why I will not use any of the results from the ASMI study on long toss specifically related to torso or PAS kinematics as their measurement system is incapable of accurately differentiating these types of outputs given their inherent limitations.



In the External/Internal Rotation section of Dillman et al. (1993) they address another limitation of their measurement methodology: “The exact contribution to total external rotation by each of the shoulder components of glenohumeral, scapulothoracic, and trunk hyperextension was not quantified in this study.”  That was potentially a problem in 1993 based on restrictions in camera resolution at the time.  However, camera technology has advanced sufficiently to increase the number of markers significantly to better measure the individual contribution of hyperextension of the spine and/or scapulothoracic rotations relative to glenohumeral rotation.  In Section 9.4.2 in Biomechanics of Pitching by Zheng et al. 2004 they write: “Because of how it is calculated, this maximum external rotation is actually a combination of glenohumeral rotation, sternoclavicular motion, and extension of the spine.”  The fact that ASMI has not increased the number of markers on the torso or PAS humerus leaves room for doubt regarding the absolute accuracy of these kinematic outputs.


The biggest problem with the limitation of measuring torso kinematics with the previously described 4 markers is the fact that the 2 shoulder markers are used to define the x axis of the shoulder local coordinate system (LCS) first.  As evidenced by the photos and videos above, this can become very problematic given the extreme mobility differences in pitchers.  By not accounting for scap loading differences, the shoulder line vector inaccurately reflects the torso kinematics and the resulting x axis of the local LCS at the shoulder.  While subsequent Y axis and Z axis calculations for the LCS ensure orthogonality, it is still done after 1st defining this X axis.  So it is easy to see how this modeling error gets propagated through all subsequent kinematics calculations.

2.) Marker protocol for PAS kinematics:  as was discussed in the rigid body motion tracking page, at least 3 non-collinear markers are necessary to track the position and orientation of a body segment.  It is impossible to accurately calculate the orientation of a body segment with three degrees of freedom (DOF) without 3 non-collinear markers. The use of 2 markers allows for the calculation of 2 DOF, but more information is necessary to provide accurate orientation information for the 3rd DOF.  Details on the rationale for their marker set is discussed in Section Marker Set-up for Baseball Pitching of the Biomechanics of Pitching paper, “A third point is needed to determine the rotation about the long axis of the upper arm or thighSince both the elbow and knee are considered as hinge joints, the rotation about the upper arm axis or thigh axis can be determined by knowing the position of the wrist or ankle if three markers on these joints.”  

However, as was discussed in detail in the End Effector Mobility page, the human arm has 7 DOF, so there is redundancy with an extra DOF when analyzing 6 DOF end effector mobility.  One way to see this is to place your hand firmly on a desk or wall and straighten your arm out.  Now without moving your hand off of the desk or wall and while also keeping the shoulder center in the same location, try and move your elbow around. Despite the motion constraints at the hand and shoulder, we still have the ability to rotate the elbow quite a bit.  This shows that without moving the wrist, which ASMI is using the 2 wrist markers to calculate humeral rotations, we can still observe humeral internal/external rotations at the elbow.  Specifically, the orientations of the lateral and medial humeral epicondyles will rotate in internal/external rotation with both the shoulder and wrist fixed. This ultimately changes the orientation of the humeroulnar joint (elbow hinge joint) with regards to the acromion and wrist markers, which have not moved at all.

Details on why they chose this marker set was discussed in the ASMI methodologies page regarding Section Marker Set-up for Baseball Pitching of the Biomechanics of Pitching paper.  However, this test just shows that there can be significant humeral rotations that can be missed with this marker protocol, as the acromion and 2 wrist markers would not move given this test condition.  The lateral humeral epicondyle would rotate with the humeral movement; this would only slightly affect the ASMI humeral and forearm vectors as only one of the 2 marker points is moving for each vector defined segment.

The modeling assumption employed by ASMI of using only 2 markers on the PAS humerus and using two PAS wrist markers to estimate humeral rotations significantly affects their system gauge R&R.  If one were to test the humeral external/internal rotation gauge R&R, they would find that the amount of variability induced in the kinematics by the measurement protocol itself would preclude the use of this protocol from any scientific study.  Now to be fair, during the critical arm acceleration phase when the humerus goes from maximum external rotation (MER) to maximum internal rotation (MIR), the humerus is primarily undergoing humeral internal rotation so the measurement may serve as a fairly good approximation.  However, pitchers will initiate forearm pronation at different times and at different rates during this phase, and the ASMI methodology cannot accurately separate these different long axis rotations.  Again, this is problematic if a scientist is using PAS long axis angular velocities for player profiling or is using the data for further kinetic analysis.


From the Methods section of the Dillman et al. 1993 paper, they write “Although this technique may have less accuracy than the method of fixing three rigid markers for each arm segment, it was necessary since the resolution of the system did not allow both three markers per segment and total body analysis. That was potentially a problem in 1993 based on restrictions in camera resolution at the time.  However, camera technology has advanced sufficiently to increase the number of markers significantly to better measure the shoulder kinematics using 3 non-collinear markers on the humerus.  Additional markers on a body segment are often used to increase the accuracy of the reconstructed position and orientation of the body segment.  Zheng et al. 2004 states that the use of additional markers requires “more cameras in order to see each marker by at least two cameras at any time.”  Yet ASMI has gone from 4 camera systems to 6 camera systems to even 8 camera systems as detailed in the ASMI Methodologies page, yet they have not changed their marker protocol during that time. 

One of the biggest concerns I have with the ASMI measurement methodology is they only use 2 markers on the PAS humerus and try to estimate humeral external/internal rotation through wrist markers.  This severely limits the gauge R&R measurement capabilities of the pitching arm side (PAS) kinematics, and brings into question the absolute accuracy of the ASMI internal/external shoulder angular displacements and subsequent angular velocity calculations.  In the Dillman et al. (1993) paper, they claimed that the angular position of a segment could still be determined with an accuracy of 5° despite their methodology limitations.  However, the 7 DOF arm demonstration earlier with motion constraints at both ends produced significant humeral rotations.  This error source can be very high for pitchers that initiate movements with their hand.  For example, when pitchers break their hands from their gloves, many will take the ball away and show it to second base which uses more pronation of the forearm than humeral internal rotation.  If they also simultaneously abduct the humerus during takeaway, we see players that combine internal rotation of the humerus and pronation of the forearm during takeaway to varying degrees.  This is a condition that the ASMI measurement methodology will not be able to handle accurately, which is a big concern of mine as this exact scenario puts both segments of the PAS into an extremely internally rotated situation prior to being laid back in maximum external rotation (MER).

Given the importance of shoulder kinematics in the biomechanics of pitching, I personally believe is using at a minimum three markers for PAS humeral kinematics.  Unfortunately, almost all pitching biomechanics studies use a similar methodology as discussed in the Pitching Biomechanics – Shoulder Kinematics page.  The Standardization and Terminology Committe (STC) of the International Society of Biomechanics (ISB) proposed a definition of a joint coordinate system (JCS) for the shoulder, elbow, wrist, and hand as published by Wu et al. 2005.  The International Shoulder Group (ISG) supports the efforts of ISB and advocate 3 markers on the humerus for measuring 3 DOF humeral rotations as shown in the figure below.  This is the same protocol used by Chu 2012 for a pitching biomechanics dissertation.  wu

3.)  Angular velocity calculation – this is a mathematical issue that is not specific to ASMI’s methodologies.  Rather, it is very common in biomechanics research studies.  As was discussed in the Angular Velocity page, there are a number of different methods used to calculate angular velocities, yet they all produce different results.  McGhee et al. 2000 published a paper that provides all of the details necessary to derive body angular rates from Euler angle rates.  The mathematical steps necessary to do this are given in the Body Angular Rate to Euler Rate Transformation page.

This was the step that I described previously when I obtained my Eureka! moment at Callaway Golf.  To define kinematics, I typically use a kinematic chain philosophy, where distal segment rotations are calculated relative to proximal segments.  I have typically used Euler/Cardan angles or quaternions to specify orientation information.  I will alos use direction cosine matrix algorithms to work back and forth between the different orientation formats as necessary for filtering and/or orientation calculations.

I would then calculate angular velocities using a time differential of the angular kinematics, similar to ASMI or any other biomechanics study.  I had a thorough understanding of working with direction cosine matrices as well as quaternions and specifying orientations in any desired format.  I also am very familiar with the effect of the order of rotations on Euler angle rate sequences.  That is one of the downfalls of working with orientation information as the order of rotations will affect the angular kinematic outputs.  Conversely, others like ASMI will use projection angles to calculate angular kinematics.  As the angular kinematics will change dependent upon the rotational format used, the subsequent angular velocity calculation will change even more as it is the time differentiation of the angular kinematics.

That is one of the reasons why comparing angular velocity rates in research studies is so difficult.  I ran into this problem when trying to compare golf shaft angular rates for closure rate (about long axis of shaft) and release rate (about axis perpendicular to shaft providing swing plane) between a motion capture system and an inertial sensor system.  The inertial sensor system directly measures angular rates for the three axes inside the shaft in local or body axes.  The motion capture system utilized a processing algorithm to calculate the angular rates as specified above which were Euler angle rates, but these rates are not specified in body or local coordinate systems.  So directly comparing the two results is not possible, as we are essentially comparing apples to oranges.

The paper by McGhee et al. 2000 provides the steps necessary to derive the body angular rates for the motion capture system from the calculated Euler angle rates.  This relatively simple transformation algorithm, allows direct comparison between the two different measurement systems.  Having used this methodology to verify measurement agreement between two different measurement systems, I now use it for all of my angular velocity calculations, as rates specified in local body coordinate systems makes much more intuitive sense anyways.  That way I can directly compare angular velocities about anatomical axes for both intra- and inter-subject studies and get a quantifiable comparison that is independent of the methodology used for defining the angular kinematics.

Given the limitations I just detailed, one would think that I am trying to say that ASMI should be put out of business given their methodological limitations.  Nothing could be further from the truth. Despite these limitations, ASMI is able to qualitatively analyze pitching biomechanics very well, as they have been doing for over two decades.  They can still even provide important information on qualitative torso and PAS kinematics. Their descriptions of shoulder and elbow soft tissue and bony injury mechanisms are the standard by which all other studies compare against.

Where they fall short is in their ability to provide absolute quantitative data for certain kinematic values because their underlying measurement gauge R&R is inadequate for PAS kinematics. These repeatable and reproducible absolute quantitative metrics are critical, though, if we want to progress pitching biomechanics studies from qualitative in nature to quantitative performance based studies. We have to be assured that when we measure the underlying kinematic changes between different pitchers or between same pitchers on different days, that the underlying measurement system and methodologies do not contribute to the differences.

In the Biomechanics of Pitching with Emphasis upon Shoulder Kinematics paper, Dillman et al. 1993 discussed five main goals for pitching biomechanics studies:

  1. Improving qualitative understanding of the pitching motion through high-speed videography.
  2. Developing a clinical procedure for evaluating injured pitchers.
  3. Conducting quantitative three-dimensional descriptions of upper extremity kinematics.
  4. Analysis of the resultant joint forces and torques created in pitching.
  5. Conducting appropriate cadaveric studies to assess the effects of these external loads upon the internal structures of the upper extremity.

The highlighted points are the areas where I believe there are limitation in ASMI’s quantitative analysis capabilities given the inherent limitations of their measurement methodologies. I have highlighted some of the differences between ASMI’s reported PAS kinematics and kinetics vs. other studies that use different measurement systems and methodologies in the Pitching Biomechanics – Shoulder Kinematics page.

Leave a Reply

Your email address will not be published. Required fields are marked *