Multimodal Synchronized Motion Capture, Force Plate, and Radar Dataset of the One-Legged Stand Test for Fall-Risk Assessment

@article{PhysioNet-olst-mocap-forceplate-radar-1.0,
  author = {Copeland, Daniel and Linton, Evan and Zhang, Xiang and Lugaro, Hector and Mori, Buntoku and Anthony, Brian},
  title = {{Multimodal Synchronized Motion Capture, Force Plate, and Radar Dataset of the One-Legged Stand Test for Fall-Risk Assessment}},
  journal = {{PhysioNet}},
  year = {2026},
  month = jan,
  note = {Version 1.0},
  doi = {10.13026/46hn-6b25},
  url = {https://doi.org/10.13026/46hn-6b25}
}

Abstract

We present synchronized motion capture, force plate, and processed 24 GHz Frequency-Modulated Continuous Wave (FMCW) radar recordings from 32 healthy participants (15 “young” participants aged 32 or younger, and 17 “old” participants aged 64 or older) performing the One-Legged Stand Test (OLST), a clinically validated fall risk assessment. The defining events of each of the 1,241 OLST attempts (foot-lift, start-of-stability, end-of-stability, and foot-touchdown) were labeled across data types using synchronized motion capture and force plate recordings. The dataset supports research in biomechanics, digital biomarker development, radar signal processing, and fall risk characterization. Basic MATLAB and Python utilities are provided for data loading, visualization, and event alignment.

Background

Falls are a leading cause of injury, hospitalization, and loss of independence in older adults, with substantial global public health implications. According to the World Health Organization, approximately 28–35% of individuals over the age of 64 experience at least one fall annually [1]. Importantly, early detection of increased fall risk followed by timely intervention has been shown to reduce fall incidence by up to 24% [2–4]. The One-Legged Stand Test (OLST) is a widely adopted clinical tool for assessing postural control and balance; shorter OLST durations have been associated with elevated fall risk and increased mortality [5–7]. While OLST duration alone holds significant clinical value, capturing biomechanical data during the test can yield richer insights into balance control strategies. These insights have potential applications across domains such as human biomechanics, neuromotor control, rehabilitation, robotics, and aging research.

Motion capture (MOCAP) and force plate data represent the gold standard for quantifying kinematic and kinetic responses during postural tasks. MOCAP systems provide high-resolution tracking of body segment movements in three dimensions, enabling frame-level analysis of postural sway, joint motion, and compensatory strategies [8]. Force plates complement this by offering precise measurements of ground reaction forces and center-of-pressure (COP) trajectories, which are sensitive to even subtle shifts in balance [9–11]. These systems have been used extensively to identify fall risk indicators, including asymmetries, postural instability, and loss-of-balance events [12]. However, their widespread use is limited by cost, setup complexity, and the need for trained personnel in controlled environments [7].

Emerging technologies for in-home fall risk assessment aim to extend these clinical insights into real-world settings. Contact-based wearables, such as smartwatches and pressure insoles, can support longitudinal tracking but require user compliance and technical familiarity, which can be burdensome for older adults [13]. Likewise, vision-based systems (e.g., RGB cameras) raise privacy concerns despite their technical capabilities. One promising alternative is Frequency-Modulated Continuous Wave (FMCW) radar, a non-contact sensing technology originally developed for automotive and industrial applications [14], now increasingly used in health monitoring [15–17]. Radar systems are privacy-preserving, operate without requiring wearables or direct interaction, and are capable of capturing subtle micro-Doppler signals associated with postural sway and movement dynamics.

Motivated by these challenges, we present to the PhysioNet community [18] a synchronized multimodal dataset of 32 participants (15 “young” $\leq$32y, 17 “old” $\geq$64y) performing the OLST under controlled laboratory conditions. The dataset includes labeled and synchronized recordings from three sensing modalities: marker-based MOCAP, dual ground reaction force plates, and a 24 GHz FMCW radar.

Each OLST attempt includes labeled events marking the Foot-Up (FU) and Foot-Down (FD) movements: foot-lift, start-of-stability, end-of-stability, and foot-touchdown. The dataset supports research across biomechanics, radar signal processing, digital biomarker discovery, and aging-related mobility assessment, and enables direct comparison between established and emerging technologies. Basic Python utilities are included to facilitate data access, synchronization, and visualization.

Methods

Participants

Participants were required to be physically healthy and to self-report the ability to attempt a one-legged stance without assistance. “Young” adults (ages 18 to 32) and “Old” adults (ages 64 and above) were recruited to represent distinct age cohorts (Table 1). MIT’s Institutional Review Board, the Committee on the Use of Humans as Experimental Subjects (COUHES), approved the study protocol (#1911000055), and written informed consent was obtained from all participants before data collection commenced.

Table 1: De-identified Participant Demographics and Self-Reported Balance

Gender coded as 1 = Male, 2 = Female.

Instrumentation

Data were collected in a laboratory environment using three time-synchronized systems:

1. MOCAP: A Qualisys system (Qualisys, Inc., Göteborg, Sweden) tracked participants’ movements at 100 Hz. The system consists of twelve infrared cameras arranged around the room. Eighteen reflective markers were affixed to the participants’ major joints (Fig. 3 and Fig. 4), following a standard full-body configuration commonly used in balance and postural control studies [19, 20]. This setup enables comprehensive biomechanical analysis, including the capture of gross motor movements, joint trajectories, and center of mass displacement.
2. Force Plates: Two force plates (Bertec Corporation, Columbus, OH, USA), one under each foot, recorded ground reaction forces at 1200 Hz to measure postural sway and detect foot-lift and foot-touchdown events. The force plates were directly integrated with the MOCAP system and shared a reference start time for each capture.
3. FMCW Radar: A 4-channel 24 GHz FMCW radar (DemoRad, Analog Devices, Inc., Wilmington, MA, USA) captured non-contact range and Doppler data. The system features one transmit and four receive antennas. For this study, radar data was collected at a sampling frequency of 1 MHz, with each upchirp recording 256 samples over a 200 MHz bandwidth (24.1–24.3 GHz). Each frame consisted of 128 upchirps, with an upchirp duration of 256 µs and a total chirp repetition interval of 284 µs. This configuration yields an effective frame rate of approximately 27.6 Hz. The radar was selected for its high spatial resolution and low-power, non-ionizing emissions compliant with international safety standards [21].

Based on the radar’s 14-degree vertical field of view, the device was positioned at a height of 1 m and a distance of 5 m in front of the subject to capture the whole‐body stance. The MOCAP/force-plate system was calibrated to a common global coordinate frame before each session. Each radar capture consisted of 1000 frames (approximately 37 seconds). Each MOCAP/force-plate capture was 40 seconds. A linear actuator, with a MOCAP tracker and a radar-reflective plate, moved forwards and backwards at the beginning and end of each capture to time-synchronize the multi-modal data (Fig. 1).

Figure 1: Hardware setup: The FMCW radar was positioned at a height of 1m and a distance of 5m from the participant, who is standing with one foot on each force plate. A time-sync system consisting of a linear actuator instrumented with a radar-reflective plate and a MOCAP marker. (1) The time-sync system’s linear actuator was cycled, moved forward and backward, at the beginning of the capture in Position A. (2) The time-sync system was moved to Position B, out of the path of the radar, and then the balance test was performed. (3) After the balance test was completed, the time-sync system was moved back to Position A, in the radar’s path. (4) The time-sync system’s linear actuator was then cycled a second time, before the end of the capture. The actuator cycles generate two distinct signals, at the beginning and end of the capture, in both the radar and MOCAP data, enabling time synchronization.

OLST Protocol

Upon arrival, participants provided informed consent and completed a brief warm‐up to minimize injury risk. MOCAP markers were placed on major joints, and participants stood on force plates throughout testing.

Both “Young” and “Old” cohorts performed three short (4-second) OLST captures on each leg. Additionally, at the end of the data collection, the “Old” cohort performed three long (20-second) OLST captures on each leg (Fig. 2).

If a capture was suspected of being compromised (e.g., stepping off the force plate or excessive arm movement), an additional capture was performed to ensure three usable trials per condition. Confirmed compromised trials were removed.

Each short OLST capture consisted of three 4-second attempt windows, for a total of nine attempt windows on each side. This short duration was selected based on prior findings suggesting that the first 3-5 seconds of the OLST are particularly informative for distinguishing postural stability and neuromuscular control, especially in older adults [22]. Audio and visual cues marked the start and end of each window.

During each 4‐second window, participants performed a variation of the OLST pose used in prior literature [23], known as a yoga tree pose, by raising one foot and pressing its sole against the inner thigh (or calf) of the standing leg (Fig. 3). If the lifted foot touched down before the 4-second window elapsed, they were instructed to re-attempt the OLST until the window closed. Similarly, during the long OLST captures, if the lifted foot touched down before the 20-second attempt window elapsed, they were instructed to re-attempt the OLST until the window closed.

Consequently, participants with greater balance ability typically produced fewer, longer attempts, whereas participants with poorer balance produced a larger number of shorter attempts within the same capture window. All attempts were retained and annotated individually.

In all trials, arms were held overhead and remained raised for the full capture period.

Figure 2: Visualizes relationships between Captures, OLST Attempts Windows, OLST Attempts, and Actuator Interference. Each green box indicates an OLST attempt window, during which participants were instructed to perform the One-Legged Stand Test (OLST). The start and end of each attempt window were marked by synchronized audio and visual cues. Gray regions indicate periods of time sync actuator interference. If a participant stepped down early, they were encouraged to reattempt within the same window. Red blocks represent individual OLST attempts as illustrated in Fig. 5. The thin red block may correspond to an incomplete OLST attempt in which the participant never achieved a stable one-legged stance (i.e., missing the Stability Phase). Additionally, attempts that abut the end actuator interference region likely do not include a Foot-Down (FD) movement, as the recording ended while the participant was still in the one-legged stance. The radar row shows continuous frame-by-frame recording across the capture, taken alongside the MOCAP/Force Plate recordings in both the Short and Long Captures. Time is shown in seconds for MOCAP/Force Plate and in frames for radar.

Figure 3: Participant performing the OLST on force plates: (a) two-leg starting position; (b) left-sided one-leg OLST position. (c) Motion‐capture marker placement (eighteen green markers) and force‐plate showing ground reaction forces (red vector).

Figure 4: Illustration of MOCAP marker locations. Anatomical positions of the 18 markers used for motion capture are shown on anterior, lateral, and posterior skeletal views. Markers were placed on bilateral wrists, elbows, shoulders, hips (anterior and posterior), knees, and ankles, as well as on the chest, belly, upper back, and lower back.

Table 2: MOCAP Marker Locations and Anatomical Descriptions

Data Description

All analyses were performed in MATLAB R2022a (MathWorks, Natick, MA), QTM 2022.3 (Qualisys AB, Gothenburg, Sweden), or Python 3.10 using SciPy, NumPy, and custom scripts developed for multimodal synchronization, signal processing, and event tagging.

Tagging Process for the OLST’s Defining Events

Synchronized force plate and MOCAP data were used to tag the defining events in each OLST attempt. These key motion events segment a complete OLST attempt into four phases: (1) Two-Leg Standing (pre-foot-lift), (2) FU (foot-lift to start-of-stability), (3) Stability Phase (start- to end-of-stability), and (4) FD (end-of-stability to foot-touchdown) (Fig. 5).

Note: In some attempts, participants never achieved stability, resulting in the absence of phase (3). In other cases, captures ended while the participant was still in the stability phase (3), so there is no FD, phase (4). In these instances, only a partial sequence (e.g., missing phase (3) or (4)) is available for analysis. More details about these edge cases are provided in the Section Interpretation notes.

Note: All participants in the dataset completed OLST trials on both the left and right sides for all prescribed short- and long-duration captures. No sides were omitted due to safety concerns or participant ability.

Figure 5: Sequential representation of a complete OLST attempt. Participants transition from a starting two-legged standing pose to a one-legged stance, followed by a return to the initial pose. Synchronized MOCAP and force plate data were used to identify key temporal events: foot-lift marks the initiation of the Foot-Up (FU) phase, start-of-stability indicates balance achievement and the beginning of the stability phase, end-of-stability designates the end of the stability phase and the start of Foot-Down (FD), and foot-touchdown signals the foot’s return to the force plate at the end of FD and a return to two-legged standing.

The tagging process involved the following three main steps: (1) detection of foot-lift and foot-touchdown events using force plate data, (2) estimation of start and end-of-stability using MOCAP-derived knee angle analysis, and (3) manual verification of all events.

Step 1: Force Plate-Based Detection of Foot-Lift and Foot-Touchdown Events

1. Annotation: Each force plate was assigned to either the lifted or the standing foot based on trial metadata.
2. Foot Lift Identification: Dynamic thresholding was applied to the lifted leg’s force data to detect transitions from nonzero to zero force (foot-lift) and zero to nonzero force (foot-touchdown).
3. Event Validation: Detected events were validated to ensure physiologically plausible sequences (e.g., each foot-touchdown followed a preceding foot-lift).
4. Manual Review of foot-lift and foot-touchdown: Events were manually reviewed in QTM Software using the synchronized MOCAP trajectories and force data to ensure accurate timing.

Step 2: MOCAP-Based Detection of Stability Phases

1. Knee Angle Calculation: Time series of lifted leg knee flexion-extension angles were computed using 3D MOCAP marker coordinates to define the stability phase (Fig. 5).
2. Stability Phase Identification: Dynamic thresholding with hysteresis was applied to the derivative of the lifted leg knee angle, marking start-of-stability (end of FU) when the derivative approached zero, and end-of-stability (start of FD) when it began to increase.

Step 3: Manual Verification

Annotated events from force plate and MOCAP sources were cross-validated and refined through manual inspection of the RGB video data to correct edge cases and remove ambiguous trials. These finalized timestamps served as synchronized ground truth for aligning radar data with corresponding motion events.

Radar Data Processing

With event timings precisely defined, we then processed the raw FMCW radar data to generate Range-Doppler Maps (RDMs), which contain time-resolved spatial and velocity features. The raw FMCW radar data was processed and tagged through the following pipeline (Fig. 6):

RDM Generation Pipeline

1. Signal Mixing: Transmit (Tx) and receive (Rx) chirps were mixed to produce an intermediate‐frequency (IF) signal whose beat frequency encodes range information.
2. Phase & Amplitude Calibration: Static gain and phase corrections were applied to each IF channel to compensate for Rx-chain imbalances.
3. Channel Summation: Calibrated IF signals from the four Rx antennas were coherently summed to boost signal-to-noise ratio.
4. Windowing: A Blackman–Harris window was applied across fast-time (within a chirp) samples to reduce sidelobes, unwanted spectral leakage that can obscure weaker signals.
5. Range-Doppler Map (RDM) Generation: Two successive Fast Fourier Transforms (FFTs) were performed across fast-time samples and slow-time (consecutive chirps within a frame) samples to generate 2D RDMs indexed by range and velocity.
6. Target Region Cropping: Two separate RDMs were generated by cropping along both range and velocity dimensions to suppress background clutter. One RDM was cropped to range bins corresponding to the participant location (4 to 6 meters) and velocity bins within physiologically plausible velocities (–5 to 5 m/s). A second RDM was cropped to range bins corresponding to the actuator location (1 to 3 meters) and velocity bins similar to the first RDM.

The resulting RDM outputs were further processed to detect the time-sync system’s actuator signal for synchronization with the MOCAP/force-plate data. In addition to actuator movement detection, the RDMs could also be used for wobble analysis to quantify participant stability or be further preprocessed for machine learning.

Figure 6: Range-Doppler Map (RDM) Generation: (1) Signal Mixing of Transmission (Tx) and Receive (Rx); (2) Phase & Amplitude Calibration; (3) Channel Summation; (4) Windowing; (5) Range-Doppler Map (RDM) Generation via Successive Fast Fourier Transform (FFT); (6) Target Region Cropping.

Data Records and Organization

The dataset is organized into four top‐level directories:

Metadata/
+-- FMCWRadar_Settings.xlsx
+-- ForcePlate_Settings.xlsx
+-- MOCAP_Markers_Locations.csv
+-- MOCAP_Settings.xlsx
+-- OLST_Attempts.csv 
+-- Participant_Demographics.csv

OLST_QTM/
+-- AIM models/
+-- Calibrations/
+-- Data/
|   +-- 01/
|   |   +-- 01_MNTRL_MC_V1.qtm
|   |   +-- 01_MNTRL_MC_V2.qtm
|   |   +-- …
|   +-- 02/
|   +-- …  
+-- Messages/
+-- Settings.qtmproj

Processed/
+-- Radar_RDMs/
    +-- 01/
    |   +-- 01_MNTRL_RR_V1.mat
    |   +-- …
    +-- 02/
    +-- …  

Raw/
+-- MOCAP/
|   +-- 01/
|   |   +-- 01_MNTRL_MC_V1.csv
|   |   +-- …
|   +-- 02/
|   +-- …  
+-- ForcePlate/
|   +-- 01/
|   |   +-- 01_MNTRL_FP_V1_left.csv
|   |   +-- 01_MNTRL_FP_V1_right.csv
|   |   +-- …
|   +-- 02/
|   +-- …  

File Naming Convention

Each capture, consisting of simultaneous, time-synchronized MOCAP, force plate, and FMCW radar recordings, was saved using the following standardized filename structure:

ParticipantID_MovementCode_Modality_VersionNumber

Each filename includes a modality code:

• RR – Radar Recording
• MC – Motion Capture
• FP – Force Plate

Table 4: Movement Code Definitions

For example, if participant “01” performed a right-sided (right foot as base) Mountain-to-Tree Short capture for the second time and the recording modality was radar, the filename would be:

01_MNTRR_RR_V2

Short trials consist of three 4-second attempt windows per capture. Long-hold trials consist of a single 20-second attempt window.

Although older participants completed long-duration OLST capture sessions (20 seconds), the annotated balance intervals (from Foot-Up to Foot-Down) were often substantially shorter, reflecting the time required to reach the position and early loss of single-leg stability.

OLST Attempt Events CSV

The synchronized MOCAP/force-plate timestamps and radar frames, which define each OLST attempt (foot-lift, start-of-stability, end-of-stability, foot-touchdown), are recorded in the file OLST_Attempts.csv.

Each of the 1,241 rows in the file OLST_Attempts.csv corresponds to a single OLST attempt and contains the four time-aligned key postural events in both Radar frames and MOCAP/force-plate time:

Unique Attempt Identifier

• OLST_attempt_id — OLST attempt’s unique identifier. A combination of the radar capture identifier and the attempt number within the capture.

Ex: 01_MNTRR_RR_V2_an2

Capture Information

• RADAR_capture — radar capture identifier.

Ex: 01_MNTRR_RR_V2

• MOCAP_Start_Time, MOCAP_End_Time — timestamps (in seconds from start of data capture) marking the interval when the time synchronization actuator was visible to both radar and MOCAP.

Ex: 7.65, 30.25

• RADAR_Start_Frame, RADAR_End_Frame — corresponding radar frame indices for the actuator visibility window.

Ex: 200, 819

• Seconds_per_Frame — the calculated time duration of a radar frame during this capture:

$\begin{equation} \texttt{Seconds\_per\_Frame} = \frac{\texttt{MOCAP\_End\_Time} - \texttt{MOCAP\_Start\_Time}}{\texttt{RADAR\_End\_Frame} - \texttt{RADAR\_Start\_Frame}} \end{equation}$

$\begin{equation} \texttt{Seconds\_per\_Frame} = \frac{30.25 - 7.65}{819 - 200} = \frac{22.60}{619} \approx 0.0365 \end{equation}$

Ex: 0.0365

Attempt Information

• an — OLST Attempt Number (an) within the capture.
  • Note: The total number of attempts within a capture will change based on the participant's balance ability. Typically, better balancers will have fewer, longer attempts than poorer balancers. Min attempts per capture in dataset = 1, Max attempts per capture in dataset = 15.

Ex: 2

• is_attempt_final — Boolean flag indicating whether the attempt was the final window in the capture

Ex: False

MOCAP/Force-plate Event Times

Units: Seconds from start of capture.

• t_foot_up — Foot-lift (start of Foot-Up).

Ex: 18.65

• t_stable — Start-of-Stability (end of Foot-Up).

Ex: 19.28

• t_break — End-of-Stability (start of Foot-Down).

Ex: 22.37

• t_end — End of the OLST attempt.

Ex: 23.35

Radar RDM Events Frame Indices

Units: Frames from radar start of capture.

• frame_foot_up, frame_stable, frame_break, frame_end — radar frame indices aligned to each respective event.

Ex: 501, 518, 603, 630

Event synchronization notes:

t_X: Absolute time in seconds from the start of the MOCAP/force-plate recordings. Used for precise temporal alignment across systems.

frame_X: Index of the radar frame in the RDM sequence, corresponding to the associated event.

Interpretation notes:

If t_stable is missing, the participant never achieved a stable knee angle during the OLST attempt.

If t_break is missing, the participant remained in a stable stance until t_end.

If t_end reflects actuator movement (rather than a true Foot-Down event), it marks the last usable frame for radar analysis in that trial.

All annotations were generated using synchronized MOCAP/force-plate signals and verified manually for consistency as described in the beginning of this section.

Technical Validation

The MOCAP/force-plate system was calibrated before each experimental session to minimize spatial error and ensure consistent marker visibility. Gaps in marker trajectories were rare and, when present, were filled manually using pattern-based gap interpolation in Qualisys Track Manager (QTM), which leverages the velocity and shape of nearby trajectories. Ground reaction forces were acquired from dual force plates that were inspected and zeroed before each session. These calibration and preprocessing steps follow best practices used in prior biomechanics data descriptor studies [24-29].

Raw marker trajectories and force plate data were visually inspected in QTM to ensure tracking consistency, correct marker labeling, and data completeness. Trials with significant dropouts or labeling errors were excluded or repeated.

Time synchronization across radar, motion capture, and force plate systems was achieved using a linear actuator signal visible in both radar and MOCAP data (Fig. 1). Frame rate for each radar capture was calculated using the time difference between actuator events at the start and end of each sequence.

Across all participants, the standard deviation of the average frame interval (Seconds_per_Frame) was 0.00059s (0.43 Hz), reflecting moderate variability across sessions, likely due to hardware or environmental factors between capture days. This variation may stem from subtle differences in USB bus timing, processor load, or ambient electromagnetic conditions on different recording days. Within individual participants, however, frame interval variability was much lower: the median within-participant standard deviation was only 0.00011s (~0.08 Hz), and all trials fell within three standard deviations of the mean.

This high within-participant consistency confirms the reliability of our actuator-based alignment procedure, which successfully identified synchronization points in both radar and MOCAP streams without requiring hardware triggers. These results support the dataset’s high temporal fidelity and its suitability for multimodal analysis of postural dynamics and cross-sensor event timing.

Usage Notes

Intended Use and Dataset Scope

This dataset is intended to support methodological and comparative research in human balance assessment, including biomechanics, radar signal processing, digital biomarker development, and aging-related mobility analysis. The synchronized motion capture, force plate, and FMCW radar recordings enable direct comparison between established laboratory-based measurements and emerging non-contact sensing approaches, as well as the development and validation of algorithms for event detection, postural stability characterization, and multimodal sensor fusion.

Limitations

Several limitations should be considered when reusing this dataset.

• First, the number and duration of annotated OLST attempts per participant vary as a function of individual balance performance, as attempts were terminated upon foot touchdown and participants were instructed to re-attempt within the same window; consequently, better balancers tend to have fewer, longer attempts, whereas poorer balancers have more, shorter attempts.
• Second, one capture (12_MNTRL_V2) is missing due to a data collection error and should be excluded from analysis.
• Third, long-duration (20 s) OLST captures are available only for older participants, reflecting the study’s focus on sustained balance performance in this cohort.
• Fourth, radar frame rates exhibit mild variability across participants; therefore, conversions between radar frame indices and time should use the row-specific Seconds_per_Frame value provided in the metadata.
• Finally, all data were collected under controlled laboratory conditions in physically healthy participants, which may limit generalizability to unsupervised, in-home settings or clinical populations with significant mobility impairments.

Radar Timing and Synchronization Notes

Radar captures exhibit very mild variability in frame rate across participants. Accordingly, any conversion between radar frame indices and time should use the row-specific Seconds_per_Frame value provided in the metadata to ensure accurate temporal alignment with motion capture and force plate data.

Optional Software and Visualization Resources

The .qtm files in OLST_QTM/ are provided for visualization, inspection, and reference within Qualisys Track Manager (QTM) and are not required for reuse or analysis of the dataset; all data necessary for reuse are available in the exported CSV, MAT, and metadata files.

To explore the synchronized motion capture recordings in QTM, users may:

1. Download the full OLST_QTM folder from the dataset.
2. Rename the folder AIM_models to AIM models (the underscore was required for PhysioNet upload).
3. Launch Qualisys Track Manager on a local machine.
4. Open the project by selecting the OLST_QTM folder.
5. Use the project interface to browse or play back 3D marker trajectories aligned with the force plate data.

Researchers are also encouraged to consult the companion Jupyter notebook (viz_OLST_attempt.ipynb), which demonstrates how to visualize OLST attempts and integrate metadata from OLST_Attempts.csv

Release Notes

Version 1.0.0: Initial release of the dataset.

Ethics

The study protocol was approved by MIT’s Institutional Review Board, the Committee on the Use of Humans as Experimental Subjects (COUHES), under protocol number #1911000055. Written informed consent was obtained from all participants prior to participation. All data have been de-identified in accordance with HIPAA Safe Harbor guidelines prior to public release.

Acknowledgements

This work was funded by Sekisui House Ltd., Osaka, Japan, and approved under MIT’s Committee on the Use of Humans as Experimental Subjects (COUHES) protocol #1911000055. The authors thank all the study participants for their time and effort, and acknowledge the support of the MIT Center for Clinical and Translational Research staff.

Conflicts of Interest

The authors declare no competing interests.

References

1. World Health Organization. WHO global report on falls prevention in older age. Geneva: World Health Organization; 2008. Available from: https://www.who.int/publications/i/item/9789241563536
2. Gillespie LD, Robertson MC, Gillespie WJ, Sherrington C, Gates S, Clemson L, Lamb SE. Interventions for preventing falls in older people living in the community. Cochrane Database Syst Rev. 2012;9:CD007146. doi:10.1002/14651858.CD007146.pub3
3. Ong MF, Soh KL, Saimon R, Wai MW, Mortell M, Soh KG. Fall prevention education to reduce fall risk among community-dwelling older persons: a systematic review. J Nurs Manag. 2021;29:2674–2688. doi:10.1111/jonm.13434
4. Ott LD. The impact of implementing a fall prevention educational session for community-dwelling physical therapy patients. Nurs Open. 2018;5:567–574. doi:10.1002/nop2.165
5. Araujo CG, de Souza Silva CG, Laukkanen JA, Singh MF, Kunutsor SK, Myers J, Franca JF, Castro CL. Successful 10-second one-legged stance performance predicts survival in middle-aged and older individuals. Br J Sports Med. 2022;56:975–980. doi:10.1136/bjsports-2021-105360
6. Springer BA, Marin R, Cyhan T, Roberts H, Gill NW. Normative values for the unipedal stance test with eyes open and closed. J Geriatr Phys Ther. 2007;30:8–15. doi:10.1519/00139143-200704000-00003
7. Chen B, Liu P, Xiao F, Liu Z, Wang Y. Review of upright balance assessment based on force plates. Int J Environ Res Public Health. 2021;18:2696. doi:10.3390/ijerph18052696
8. Maudsley-Barton S, Yap MH. Objective falls risk assessment using markerless motion capture and representational machine learning. Sensors. 2024;24:4593. doi:10.3390/s24144593
9. Wikstrom EA, Tillman MD, Smith AN, Borsa PA. A new force plate technology measure of dynamic postural stability: the Dynamic Postural Stability Index. J Athl Train. 2005;40:305–309.
10. Quijoux F, Nicolaï A, Chairi I, Bargiotas I, Ricard D, Yelnik A, Oudre L, Bertin-Hugault F, Vidal PP, Vayatis N, Buffat S, Audiffren J. A review of center-of-pressure variables to quantify standing balance in elderly people: algorithms and open-access code. Physiol Rep. 2021;9:e15067. doi:10.14814/phy2.15067
11. Blaszczyk JW, Orawiec R. Assessment of postural control in patients with Parkinson’s disease: sway ratio analysis. Hum Mov Sci. 2011;30:396–404. doi:10.1016/j.humov.2010.07.017
12. Eichler N, Raz S, Toledano-Shubi A, Livne D, Shimshoni I, Hel-Or H. Automatic and efficient fall risk assessment based on machine learning. Sensors. 2022;22:1557. doi:10.3390/s22041557
13. Patel S, Park H, Bonato P, Chan L, Rodgers M. A review of wearable sensors and systems with application in rehabilitation. J Neuroeng Rehabil. 2012;9:21. doi:10.1186/1743-0003-9-21
14. Srivastav A, Mandal S. Radars for autonomous driving: a review of deep learning methods and challenges. IEEE Access. 2023;11:97147–97168. doi:10.1109/ACCESS.2023.3312382
15. Ni X, He Y, Jing X. Survey of deep learning-based human activity recognition in radar. Remote Sens. 2019;11:1068. doi:10.3390/rs11091068
16. van Berlo B, Elkelany A, Ozcelebi T, Meratnia N. Millimeter-wave sensing: a review of application pipelines and building blocks. IEEE Sens J. 2021;21:10332–10368. doi:10.1109/JSEN.2021.3057450
17. Mandischer N, Koop I, Granich A, Heberling D, Corves B. Radar tracker for human legs based on geometric and intensity features. In: Proc 29th Eur Signal Process Conf (EUSIPCO); 2021. p. 1521–1525. doi:10.23919/EUSIPCO54536.2021.9616134
18. Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PC, Mark RG, Mietus JE, Moody GB, Peng CK, Stanley HE. PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals. Circulation. 2000;101:e215–e220. doi:10.1161/01.CIR.101.23.e215
19. Winter DA. Human balance and posture control during standing and walking. Gait Posture. 1995;3:193–214. doi:10.1016/0966-6362(96)82849-9
20. Mancini M, Horak FB. The clinical use of balance measures for screening and diagnosis of balance disorders: a systematic review. J Neuroeng Rehabil. 2010;7:1–21.
21. Fereidouni F. Human health risk assessment of 4–12 GHz radar waves using CST Studio Suite software. J Biomed Phys Eng. 2022;12.
22. Jonsson E, Seiger Å, Hirschfeld H. One-leg stance in healthy young and elderly adults: a measure of postural steadiness? Clin Biomech. 2004;19:688–694. doi:10.1016/j.clinbiomech.2004.04.002
23. Rohof B, Betsch M, Rath B, Tingart M, Quack V. The Nintendo® Wii Fit Balance Board can be used as a portable and low-cost posturography system with good agreement compared to established systems. Eur J Med Res. 2020;25:44. doi:10.1186/s40001-020-00445-y
24. Fukuchi RK, Fukuchi CA, Duarte M. A public dataset of running biomechanics and the effects of running speed on lower extremity kinematics and kinetics. PeerJ. 2017;5:e3298. doi:10.7717/peerj.3298
25. Schreiber C, Moissenet F. A multimodal dataset of human gait at different walking speeds established on injury-free adult participants. Sci Data. 2019;6:111. doi:10.1038/s41597-019-0124-4
26. Moreira L, Figueiredo J, Fonseca P, Vilas-Boas JP, Santos CP. Lower limb kinematic, kinetic, and EMG data from young healthy humans during walking at controlled speeds. Sci Data. 2021;8:103. doi:10.1038/s41597-021-00881-3
27. van der Zee T, Mundinger EM, Kuo AD. A biomechanics dataset of healthy human walking at various speeds, step lengths, and step widths. Sci Data. 2022;9:704. doi:10.1038/s41597-022-01817-1
28. Piorek M, Josiński H, Michalczuk A, Świtoński A, Szczęsna A. Quaternions and joint angles in an analysis of local stability of gait for different walking speeds and treadmill slopes. Inf Sci. 2017;384:263–280. doi:10.1016/j.ins.2016.08.069
29. Wang Y, Mei Q, Han J, Jiao X, Yang X, Liew BXW, Fernandez J, Gu Y. Dataset of walking and running biomechanics with different step widths across different speeds. Sci Data. 2025;12:802. doi:10.1038/s41597-025-05113-6