Whole-body optical marker and ground reaction force data of healthy humans performing non-cyclic activities Authors: Mohsen Alizadeh Noghani, Sebastian Green, Edgar Bolívar-Nieto Published: 2025 ABSTRACT AbstractThe motion of the human whole-body center of mass (CoM) supplies information that is useful for controlling wearable robots and studying important aspects of human movement, such as balance. However, most datasets do not use a whole-body markerset or provide the full 3D ground reactions forces (GRFs), both of which can be crucial for accurately estimating the CoM, or mainly focus on cyclic activities such as walking. This work contains the whole-body markerset and 3D GRF measurements of 10 participants performing 14 activities, 11 of which were derived from the Berg Balance Scale test. Additionally, we provide the results of inverse kinematics and dynamics analysis, as well as estimates of the CoM position and velocity. The information in this dataset enables study of human movement in novel scenarios, with possible applications such as quantifying balance and designing controllers for powered prosthetic legs.Subject terms: Mechanical engineering, Biomedical engineering Background & Summary The human center of mass (CoM) plays a key role in quantifying human movement. For example, in static equilibrium, stability can be defined as the distance between the vertical projection of the CoM position and the boundaries of the base of support (BoS)1. Meanwhile, in dynamic scenarios, it is the combination of the CoM position and velocity, represented by the extrapolated center of mass (XCoM)2, that determines conditions for falling3. The CoM state has also been used for predictive applications, such as estimating changes in width of the next step4, and and in control of powered lower-limb prostheses5, to define stabilizing torque of a prosthetic ankle6. Strategies to estimate the CoM position and velocity can be divided into two main groups. The first uses double integration of ground reaction forces (GRFs) measured by force plates and subsequent filtering to compensate for the drift caused by integration of the noisy signal7–9. The second approach models the human as a chain of rigid bodies and estimates the overall CoM by calculating the kinematics of the CoM for one or multiple segments10–12. Although combining a whole-body markerset with anthropometric tables13–15 to compute individual segment CoMs can result in more accurate estimates, a common simplification is using only the position of markers placed on the pelvis16–19. It should be noted that this approximation of the 3D CoM trajectory would induce higher errors in movements involving significant displacements of body segments, compared to those performed in or near an upright body posture12,20. Previous work on comparing methods for estimation of CoM kinematics has mainly considered 2D motion, particularly during walking, with a small number of activities studied in 3D. Lafond et al.10 examined CoM displacement in the anteroposterior (AP) and mediolateral (ML) directions for standing tasks, such as standing on one leg. Havens et al.19 characterized gait stability while turning via XCoM trajectories in 2D, while Buurke et al.9 compared estimation via markers and GRFs during treadmill walking in the ML direction. In 3D, marker- and GRF-based methods of estimating CoM kinematics have been compared during squatting or lifting movements8, in various modes of gait (e.g., skipping) on a treadmill12,21,22 and overground23, and during recovery from unexpected slips in older adults24. Considering the limitations of these works (i.e., an emphasis on cyclic motion and/or a lack of 3D CoM information), a dataset containing measurements of a whole-body markerset and 3D GRFs in a wider range of activities and for longer durations could enable more effective comparison of different CoM estimation methods and study of the CoM dynamics, particularly for applications in control of powered prosthetic legs. However, the existing datasets have been mainly focused on level walking and running25–31. Meanwhile, some cover a wider range of activities, but only use a lower-limb markerset32–34, whereas others do not report the GRFs35 or only do so partially36. Motivated by the gaps present in the current datasets, we collected 3D GRF measurements and data of a whole-body markerset for 10 subjects performing 14 non-cyclic activities, 11 of which were derived from the tasks in the Berg Balance Scale test (BBS)37. This dataset extends the work by Liang et al.38 with a larger number of tasks derived from the BBS, and particularly ones that focus on different configurations of the lower limbs, which is of interest in applications that utilize the CoM for controlling powered lower limb prostheses. Additionally, for reproducibility of the results, we present the MATLAB scripts used for calculating inverse kinematics, inverse dynamics, and CoM position and velocity in OpenSim11. We have previously used this data to investigate estimating the CoM kinematics using pelvis and whole-body markersets as well as the GRFs20, and to predict the CoM position in a future horizon for different horizon lengths and under various assumptions for the CoM acceleration39. Methods Participants We collected data of ten healthy able-bodied participants (n = 10, mean ± SD: age 22.9 ± 3.5 years, height 1.73 ± 0.08 m, mass 63.66 ± 9.79 kg), including five males and five females, with no self-reported pathologies affecting balance or gait. The study protocol, including sharing the data presented in this work, was approved by the University of Notre Dame Institutional Review Board (Protocol ID: 24-01-8277) and the participants gave written informed consent to take part in the study and for their de-identified videos and photos to be used in publications. Participant Preparation Each participant wore fitted elastic shirt and shorts, along with low-cut socks, and fifty-seven markers were placed on them according to the Biomech (57) whole-body markerset40 (Fig. 1). The participants wore the socks to mitigate the effects of different shoe geometries and materials on human biomechanics. (a) The anterior and posterior views of the Biomech (57) markerset. The calibration markers are highlighted in red. (b) A participant fitted with the markerset, standing in an A-pose on the instrumented walkway. The participant provided consent for publication of this image. Instrumentation Six Kistler 9260AA6 force plates (Kistler Instrumente AG, Winterthur, Switzerland) (Fig. 2) recorded center of pressure (CoP) and 3D GRFs at 1000 Hz, while a twelve-camera motion capture system (PrimeX22, NaturalPoint Inc., OR, USA) collected the marker positions at 200 Hz. The six instrumented force plates, together with six non-instrumented plates, formed a walkway (Fig. 2) (length: 2.84 m, width: 1.07 m), inside a larger motion capture volume (Fig. 3) (length: 7.19 m, width: 5.47 m, height: 2.42 m). The motion capture volume, along with the placement of cameras and force plates, is also shown in the file Video_S1_MotionCapture_Activities.mp4 in the data repository. The six force plates used during data collection (outlined in red), with their corresponding numbers in the dataset. The six non-instrumented force plates are outlined in green. As a representative example, the x5, y5, and z5 axes specify the direction and location of the local coordinate frame of force plate 5. The X, Y, and Z axes show the direction of the motion capture system frame, which is used to express the marker and force plate data. We note that the origin of the motion capture frame is outside the volume captured in this image; its location is shown in Fig. 3. The origin of the motion capture frame is at point O, with the X, Y, and Z axes defining its orientation. The numbers assigned to the five cameras (i.e., 3 to 7) are the same as those used in the TAK motion capture files and shown in the file Video_S1_MotionCapture_Activities.mp4 in the data repository. The chair was used for the Sit to Stand, Stand to Sit, and Foot on Chair activities, and is positioned as it would be for the first two. As a representative example, the x5, y5, and z5 axes specify the direction and location of the local coordinate frame of force plate 5. Experimental Procedure In this section, we describe the activities (Fig. 4) in the order they were performed on the instrumented walkway (Fig. 2) and the force plates used during each one. A hand rail was available along the length of the walkway to prevent falls if a participant lost balance; in those cases, the trial was repeated to ensure all the external forces were measured by the force plates on the ground. The activities and statistics of their durations, reported in milliseconds as mean (standard deviation) and minimum-maximum. The activities derived from the Berg Balance Scale test are marked with (*). The activities were performed at the participant’s natural pace without standardizing their timing (e.g., using a metronome). Unless specified otherwise, at the start of the activity, the participants faced the + Z axis while standing in a neutral pose with their right and left feet on force plates 1 and 2, respectively. During all the activities, foot placement was monitored so that each one was in contact with a maximum of one force plate at a time. The sitting surface of the chair, used in Sit to Stand, Stand to Sit, and Foot on Chair activities, was 454 mm above the surface of the walkway. The activities denoted by * in the list below are derived from the BBS. The file Video_S1_MotionCapture_Activities.mp4 in the data repository provides a representative example of each of the activities performed by a participant. Static (A-pose) The participant stood still outside the walkway, in an A-pose and with all of the 57 markers (Fig. 1) visible. After this, the 8 calibration markers were removed. Sit to Stand* The participant faced the −Z axis while sitting on a chair with two legs on force plate 3, one on 6, and one on the non-instrumented plate between 2 and 6. The right and left feet of the participant were on force plates 2 and 1, respectively. When instructed, they stood up at their own pace. Stand to Sit* The participant faced the −Z axis while standing, with their right and left feet on force plates 2 and 1, respectively. Two legs of the chair were on force plate 3, one on 6, and one on the non-instrumented plate between 2 and 6. When instructed, they sat down at their own pace. Foot on Chair* A chair was placed on force plate 3. The participant started from their neutral pose standing on force plates 1 and 2. They placed their right foot on the chair, returned to their original position, and then replicated a similar movement with their left leg without touching the chair (i.e., the right foot touched the chair, while the left foot did not). This was repeated until their right foot had contacted the chair for four times. Stride The participant took one stride on the walkway along the +Z axis. Their right foot was in contact with force plates 1 (at the start), 3 (at the first heel strike), and 4 (at the second heel strike). For their left foot, the contact sequence was with force plates 2, 6, and 5. Look Behind* The participant looked directly behind by turning their upper body, first looking over their left shoulder, and then over their right. Feet Together* The participant’s right and left feet were placed as close together as possible along the X-axis, while remaining on individual force plates. They maintained this position for at least 15 seconds. Eyes Closed* The participants stood still in their neutral pose and with their eyes closed for 10 seconds. Turn* The participants turned around 180 degrees (i.e., to face the −Z axis) and then returned to their original pose. This activity was performed on force plates 1, 2, and 3. The sequence of force plate contacts of the right foot was 1-3-2-3-1, while the left foot contacted force plates 2-1-2. Participants were instructed to follow this order so that each force plate was contacted by no more than one foot at a time; their motion was also monitored to ensure no foot touched two plates simultaneously. Tandem Feet* The participants faced the +Z axis while their right foot was on force plate 1 and their left foot was on force plate 3. Their feet were placed as close together as possible along the along the Z-axis (i.e., in tandem), without touching each other, and remained on separate force plates. They maintained this position for at least 15 seconds. One Leg* Participants lifted one leg until the thigh was approximately horizontal—or lower if maintaining balance was difficult—and held the position for at least 10 seconds before returning to the starting posture. Half of the participants were randomly assigned to lift their dominant leg, and the other half to their non-dominant leg. The dominant leg was determined by their choice to kick a ball. Squat The participants moved to a squatting pose and maintained it for at least 8 seconds. Then, they returned to their starting neutral pose. Reach Forward* The participants lifted their arms, reached forward until they achieved their maximum reach, and then returned to their starting pose. Both feet maintained contact with the ground during this activity. Object Pickup* The participants reached down and picked up an object with negligible mass placed on force plate 3. Shoe Lace The participant placed their right foot on force plate 3. They reached down and held their hands close to that foot without making contact with it. They maintained this position for at least 8 seconds, and then returned to their starting pose. Motion Capture Post-processing The markers were labeled and the gaps in their trajectories were filled in Motive 3.0.3 using linear interpolation. During extended gaps, particularly those caused by self-occlusion (e.g., RIAS in the Shoe Lace activity), linear interpolation did not produce realistic trajectories; in such cases, we used the “Pattern-Based” interpolator, which uses the trajectories of other, visible markers, assuming they are located on the same rigid body as the one missing41. After filling all the gaps of all the markers, their trajectories were low-pass filtered at 6 Hz (zero-lag, 4th degree Butterworth filter41) and the marker and force plate data were exported to the C3D format in the coordinate frame of the motion capture system. We note that the force plate data was not filtered in Motive. Data Processing Processing was performed using MATLAB 2023b and OpenSim 4.4. We defined a MATLAB class (trial_class), which uses OpenSim’s MATLAB API as well as modified versions of MoCapTools42 and EpicToolbox43 libraries, and contains Methods for processing the data and Properties to store the results. The scripts used during the processing steps as well as the main Methods and Properties of trial_class are described in Table 1. We customized a full-body OpenSim model44 by adding a head-and-neck segment, connected to the torso with a ball-and-socket joint, and computed the masses and moments of inertia according to anthropometric tables by Dumas et al.45. Table 1The MATLAB scripts and the primary Methods and Properties of the trial_class used for processing and storing the data of each activity.NameTypeDescriptionprocess_trial_data.mMATLAB scriptContains the processing steps for each repeat of the activitytrial_classConstructorSets the name of the activity and the path to the .c3d fileset_pathsMethodSets the paths of the .xml files and the outputsinit_forceplate_dataMethodReads the force plate and marker data from the .c3d fileprocess_force_data.mMATLAB scriptSets the force plate contacts and the external forces exerted on the right and left feetupdate_forcestruct_dataMethodUpdates the force data structure with the right and left feet forcesexport_trc_mot_filesMethodWrites the marker data to a .trc file, and the updated force data to a .mot fileosim_scaleMethodScales the model according to the data of the Static activityosim_ikMethodPerforms inverse kinematics to obtain the joint angles of the OpenSim modelosim_comMethodPerforms analysis to determine the segmental and whole-body CoM positions and velocities of the modelosim_idMethodPerforms inverse dynamics to obtain the joint moments of the modelplot_resultsMethodPlots example results from the IK, CoM, and ID analysesik_resultsPropertyThe IK results (joint angles in degrees)id_resultsPropertyThe ID results (joint moments in N ⋅ m)com_posPropertyThe segmental and whole-body CoM positions (in m)com_velPropertyThe segmental and whole-body CoM velocities (in m/s) The MATLAB scripts and the primary Methods and Properties of the trial_class used for processing and storing the data of each activity. To process the force plate data and assign the GRFs and CoPs to the right and left feet, the C3D file of each recording was imported to MATLAB. We performed lowpass-filtering at 20 Hz using a 5th order Butterworth filter and manually labeled the contacts with force plates according to a 20 N threshold33. Then, using the visualization in Motive, we manually assigned each force plate’s measurements to the right or left foot, according to how the participants performed the trial. Because Motive does not support gap-filling of force plate data (e.g., in the case of dropped packets), we used linear interpolation for missing data in MATLAB. In the entire dataset, this only occurred for a single sample of the force plate data, in Repeat 3 of Feet Together activity of AB03. The marker and force plate data were then exported to .trc and .mot files, respectively, for subsequent analysis. The mass of each participant was determined using the force plate measurements in the Eyes Closed trial. The marker recordings of the Static trial and the participant’s mass were used to scale the OpenSim model. The RMS and maximum marker errors of the scaled models, obtained by running Setup_Scale_GUI.xml for each participant in OpenSim, were 4.21 ± 0.34 cm (range:3.59-4.69) and 9.82 ± 0.74 cm (range: 9.07-11.33), respectively. While we did not individualize the scaling setup for each participant to minimize this error, we confirmed that the scaled model did not deviate from the A-pose in the Static trial by visual inspection in OpenSim Visualizer. The scaled model was used to compute the inverse kinematics (IK), inverse dynamics (ID), and the whole-body and segmental CoM positions and velocities, all at 200 Hz. After completion of data processing, the trial_class was saved as a .mat file for each instance of the activity, resulting in 420 such files, corresponding to the 10 participants, 14 activities, and 3 repeats of each activity. Data Records The dataset is available in a figshare repository46 with the directory structure shown in Fig. 5. In Motion capture data, we present the raw TAK files for each repeat of each activity in Raw TAK.zip files; these contain the full data of the motion capture recordings before any processing. The files in the Processed TAK.zip were obtained after labeling, gap-filling and low-pass filtering of the marker trajectories, and Processed C3D.zip contain the corresponding .c3d files. The .tak files can only be opened in Motive and are not required to use this dataset. We provide them to allow other users to reproduce the processing steps (e.g., handling the gaps) by starting from the raw TAK files and obtaining processed TAK files, which can then be exported to C3D. The structure of the directories containing the key data and processing files. The folders are shown in blue. In Data processing, we provide the same .c3d as outlined above, as well as the resulting .mot and .trc files in folders named after the participant’s identifier (e.g., AB01). The class containing the marker trajectories (200 Hz) and force plate data (1000 Hz) and the IK, ID, and CoM results (200 Hz) is saved in the Results folder as a .mat file (e.g., AB01_Sit_to_stand_2.mat). The height and mass of the participants are stored in data_info.mat, and the data of all the repeats of each activity and the results of IK, ID, and CoM analysis for each subject are available in the All subjects results folder. Technical Validation The motion capture system was calibrated before the start of each participant’s session to achieve an “Exceptional” calibration quality47. To reduce the bias due to drift in the force plate readings, they were zeroed between performing the three repeats of each activity. As an additional assessment of the reliability of our data collection, we compared the results of our analysis for the Sit to Stand activity with normative data from the literature, collected by Roebroeck et al.48. The IK and ID results for the lower-limb joints (Fig. 6) show reasonable agreement with some discrepancies, which can be attributed to how the movement was performed in each study; for example, we did not standardize the starting body configuration of the participant or the duration of their movement, and used a different seat and foot placement, leading to the differences in the joint angles and moments between our results and Roebroeck’s. The lower-limb joint angles and normalized moments (by mass ⋅ height) of participant AB07 for their three repeats of the Sit to Stand activity, compared with normative data from the literature48. The shaded areas represent one standard deviation from the mean. The whole-body CoM (Fig. 7) during this activity exhibits the expected behavior; the CoM moved forward (the Z-axis) and up (the Y-axis), while experiencing little displacement in the mediolateral direction (the X-axis). This is also observed in its trajectory in the sagittal plane (Fig. 8). We note that compared to Roebroeck’s data, the CoM in our study was displaced more in the horizontal (anterior) direction, which, similar to the IK and ID analysis, can be explained by differences in the experimental protocols. The ensemble means of the CoM position of all the participants for the Sit to Stand activity, expressed in the coordinate frame of the motion capture system. The axes are defined according to Fig. 2 with the X, Y, and Z axes corresponding to the mediolateral, vertical, and anteroposterior directions, respectively. The shaded areas represent one standard deviation from the mean. The ensemble mean of the CoM displacement of all the participants for the the Sit to Stand activity compared with normative data from the literature48, starting from the sit-off event. The shaded areas represent one standard deviation from the mean. The horizontal and vertical displacements correspond to the Z and Y axes in Fig. 7, respectively. Usage Notes As an example, the plots produced after execution of process_trial_data.m for the second repeat of the Sit to Stand activity of AB01 are provided in the Example figures folder. More information about using the code is available in Readme.pdf. Acknowledgements The authors would like to thank Jingshu Peng (Tsinghua University) for her help with data collection. Author contributions Mohsen Alizadeh Noghani: Conceptualization, Methodology, Software, Formal analysis, Investigation, Data Curation, Writing - Original Draft Sebastian Green: Software, Validation, Formal analysis Edgar Bolívar-Nieto: Conceptualization, Writing - Original Draft, Supervision. Data availability The dataset is available in the figshare repository46. Code availability The code used in this work is available in the figshare repository46. In the Data processing directory (Fig. 5), the Model folder contains the OpenSim model. For each participant, Setup_Scale_GUI.xml in the Setup folder of the Static activity can be loaded for scaling via OpenSim’s GUI. In the same directory, Setup_Scale.xml is used by MATLAB’s API to scale the model. For each instance of the activities, the OpenSim configuration files for inverse kinematics (Setup_IK.xml), inverse dynamics (Setup_ID.xml and Setup_External_Forces.xml), and CoM analysis (Setup_CoM_Analysis.xml) are available in the corresponding Setup folder. The processing is performed by process_trial_data.m, which creates an instance of trial_class and executes process_force_data.m to assign the force plate data to the right and left feet. The MATLAB tools folder contains the modified versions of EpicToolbox and MoCapTools libraries as well as the definition of the trial_class. In the All subjects processing folder, process_all_subjects_all_trials.m is responsible for execution of all or specific processing scripts to produce the results of the corresponding activities. Subsequently, save_all_subjects_all_trials.m combines and saves the results for individual participants in the All subjects results folder. The processing code is also available on GitHub (https://github.com/WeRoLab/healthy-whole-body-noncyclic-dataset). Updates for compatibility with newer versions of MATLAB and OpenSim, as well as any subsequent bug fixes, will be provided in this repository. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The dataset is available in the figshare repository46. The code used in this work is available in the figshare repository46. In the Data processing directory (Fig. 5), the Model folder contains the OpenSim model. For each participant, Setup_Scale_GUI.xml in the Setup folder of the Static activity can be loaded for scaling via OpenSim’s GUI. In the same directory, Setup_Scale.xml is used by MATLAB’s API to scale the model. For each instance of the activities, the OpenSim configuration files for inverse kinematics (Setup_IK.xml), inverse dynamics (Setup_ID.xml and Setup_External_Forces.xml), and CoM analysis (Setup_CoM_Analysis.xml) are available in the corresponding Setup folder. The processing is performed by process_trial_data.m, which creates an instance of trial_class and executes process_force_data.m to assign the force plate data to the right and left feet. The MATLAB tools folder contains the modified versions of EpicToolbox and MoCapTools libraries as well as the definition of the trial_class. In the All subjects processing folder, process_all_subjects_all_trials.m is responsible for execution of all or specific processing scripts to produce the results of the corresponding activities. Subsequently, save_all_subjects_all_trials.m combines and saves the results for individual participants in the All subjects results folder. The processing code is also available on GitHub (https://github.com/WeRoLab/healthy-whole-body-noncyclic-dataset). Updates for compatibility with newer versions of MATLAB and OpenSim, as well as any subsequent bug fixes, will be provided in this repository.