A Multimodal Dataset of Full-Body Kinematics and Kinetics from Laboratory- and Wearable-Based Systems Authors: Chang June Lee, Jung Keun Lee Published: 2026 ABSTRACT Kinematic and kinetic biomechanical measurements are critical for understanding human movements. Although laboratory-based systems such as optical motion capture (OMC) and instrumented treadmills provide gold-standard accuracy, their use is confined to controlled laboratory settings. Wearable systems, including inertial motion capture (IMC) and pressure-sensing insoles, enable measurements in real-world conditions but require validation against laboratory-based references. In this study, we present a comprehensive dataset that combines laboratory- and wearable-based systems to capture synchronized full-body kinematic and kinetic measurements. Twelve healthy young adults performed static postures, treadmill walking at multiple speeds and inclines, treadmill running, and functional movement tasks, including squats, weight shifts, and squat jumps. The dataset included raw signals from four measurement systems—OMC, IMC, instrumented treadmill, and insoles—as well as from biomechanical variables derived using a biomechanical model. Synchronized multimodal signals and processed biomechanical variables support diverse applications beyond level walking, including wearable sensor validation, machine learning model training, and gait or movement analysis. Background & Summary In rehabilitation, sports science, and robotics, biomechanical measurements are vital for understanding motor function and movement patterns1–6. These biomechanical measurements can be categorized into two groups: kinematics, which describes translational and rotational motion without considering forces, and kinetics, which considers the forces and moments that generate motion7,8. Traditionally, camera-based optical motion capture (OMC) systems and force plates or instrumented treadmills are used to measure kinematic and kinetic data, respectively9–13. Camera-based OMC systems track the 3D trajectories of reflective markers attached to anatomical landmarks to calculate kinematics of segments and joints using musculoskeletal models. By contrast, force plates or instrumented treadmills measure ground reaction forces (GRFs), ground reaction moments (GRMs), and center of pressure (CoP). Additionally, by fusing kinematic and force plate measurements with anthropometric data, joint kinetic data such as joint reaction forces, net joint torques, and joint powers can be estimated using inverse dynamics calculations14–16. OMC systems and force plates have long been considered the gold standards for motion analysis because of their consistently high measurement accuracy. However, these instruments, hereafter referred to as lab-based systems—can be used only in laboratory settings using installed camera arrays or force plates, limiting their applicability outside laboratories. To address the constraints of laboratory settings, wearable systems including inertial motion capture (IMC) systems and pressure-sensing insoles have been widely adopted to enable biomechanical measurements in outdoor and real-world environments. For example, IMC systems can support ubiquitous healthcare applications, such as fall detection17, energy expenditure monitoring18, and continuous monitoring of symptom severity19. Additionally, they can also be used to track movements during sport activities, such as sprinting20, cycling21, alpine skiing22, and ski jumping23. Pressure-sensing insoles have also been used for real-time daily walking24, running25, and disease monitoring26. Among these wearable systems, IMC systems provide full-body kinematic measurements27–30 by quantifying segment kinematics (e.g., segment orientation31–33) from multiple body-worn inertial measurement units (IMUs). By contrast, pressure-sensing insoles placed inside shoes measure plantar pressure to compute vertical GRFs and plantar CoP34–36. Similar to the combination of OMC and force plates, fusing IMC-based kinematics with the pressure data of the insole enables the estimation of joint kinetics37–39. However, such wearable measurements require validation against lab-based references to ensure high measurement reliability and thereby serve as the gold standard for evaluating and validating wearable technologies. When developing artificial intelligence models based on wearable sensor data, lab-based measurements are frequently used as the ground truth for supervised learning40–44. Accordingly, acquiring measurement data from lab-based and wearable systems is a fundamental prerequisite for the development and validation of wearable technologies. Several studies have published open multimodal datasets comprising kinematic and kinetic measurements acquired using both lab-based and wearable systems45–48. The TRIPOD dataset collected by Trautmann et al.45 provides treadmill walking data from 15 healthy young adults, which were recorded using seven lower-body IMUs, a pressure-distribution treadmill, and a photoelectric gait analysis system. Lavikainen et al.46 released a multimodal gait dataset collected from 51 participants using OMC, seven lower-body IMUs, floor-embedded force plates, and a sagittal-plane video. Camargo et al.47 presented a dataset collected from 22 participants performing treadmill and overground walking, as well as stair and ramp ambulation. The data were measured using OMC, an instrumented treadmill, floor-embedded force plates, IMUs, surface electromyography, and goniometers. Grouvel et al.48 published a dataset obtained from ten asymptomatic participants using OMC, eight IMUs, pressure-sensing insoles, and three force plates. Table 1 provides an overview of five open-access multimodal datasets, including the dataset presented in this study, which integrates lab-based and wearable measurement modalities. To the best of our knowledge, aside from the datasets described above, no multimodal dataset has been reported that includes both kinematic and kinetic measurements while also incorporating both lab-based and wearable systems.Table 1Summary of publicly available multimodal datasets combining lab-based and wearable sensor systems (IMU: inertial measurement unit, OMC: optical motion capture, FP: force plate, EMG: electromyography, GON: goniometer, LB: lower-body attachment, FB: full-body attachment, N: number of participants).Ref.Task typeModalityNScope45Treadmill walkingPhotoelectric, IMU(7,LB), Pressure treadmill15Dataset for developing and validating IMU-based gait analysis algorithms46Overground walkingOMC(42,LB), IMU(7,LB), FP(3), Video51Multimodal gait dataset for kinematics, kinetics, and cross-modal validation47Treadmill/Overground walking, Stairs, RampsOMC(32,LB), IMU(4,LB), FP(5), Force treadmill, EMG(11,LB), GON(3,LB)22Comprehensive locomotion dataset for biomechanics, locomotion recognition, and assistive device development48Overground walking, Functional movementOMC(69,FB), IMU(8,LB), Insole, FP(3)10Multi-sensor dataset enabling motion capture comparison and novel gait analysis methods54Treadmill walking, Functional movementOMC(102,FB), IMU(17,FB), Force treadmill, Insole12Wearable-based full-body kinematics and kinetics research with validation against lab-based systems Table 1Summary of publicly available multimodal datasets combining lab-based and wearable sensor systems (IMU: inertial measurement unit, OMC: optical motion capture, FP: force plate, EMG: electromyography, GON: goniometer, LB: lower-body attachment, FB: full-body attachment, N: number of participants).Ref.Task typeModalityNScope45Treadmill walkingPhotoelectric, IMU(7,LB), Pressure treadmill15Dataset for developing and validating IMU-based gait analysis algorithms46Overground walkingOMC(42,LB), IMU(7,LB), FP(3), Video51Multimodal gait dataset for kinematics, kinetics, and cross-modal validation47Treadmill/Overground walking, Stairs, RampsOMC(32,LB), IMU(4,LB), FP(5), Force treadmill, EMG(11,LB), GON(3,LB)22Comprehensive locomotion dataset for biomechanics, locomotion recognition, and assistive device development48Overground walking, Functional movementOMC(69,FB), IMU(8,LB), Insole, FP(3)10Multi-sensor dataset enabling motion capture comparison and novel gait analysis methods54Treadmill walking, Functional movementOMC(102,FB), IMU(17,FB), Force treadmill, Insole12Wearable-based full-body kinematics and kinetics research with validation against lab-based systems Summary of publicly available multimodal datasets combining lab-based and wearable sensor systems (IMU: inertial measurement unit, OMC: optical motion capture, FP: force plate, EMG: electromyography, GON: goniometer, LB: lower-body attachment, FB: full-body attachment, N: number of participants). While these datasets encompass diverse sensor configurations, experimental conditions, and participant groups, wearable kinematic sensors—primarily IMUs—have predominantly focused on the lower limbs, with most aboveground trials being limited to only a few steps (depending on the number of force plates). Although lower-limb kinematics plays a critical role in gait analysis, full-body kinematics remain essential for applications such as estimating GRFs, GRMs, and joint kinetics during walking using musculoskeletal models49,50. Full-body measurements are also indispensable when studying balance control51, load carriage52, and assistive-device53, where upper-body strategies and arm swings are integral to locomotor mechanics. By contrast, the present dataset54 provides synchronized whole-body kinematic and kinetic measurements during gait and functional movements, acquired from both lab- and wearable-based systems. Specifically, 45 anatomical markers and 17 IMUs, each attached to a marker cluster composed of three reflective markers, were used to capture the three-dimensional motion of the entire body, whereas an instrumented treadmill and pressure-sensing insoles were used to measure GRF and CoP. Accordingly, this multimodal dataset provides a basis for developing algorithms for estimating kinematic (e.g., orientations and joint angles) and kinetic (e.g., GRFs and joint torques) variables from wearable sensor data, while also enabling ground truth validation of the resulting estimates. In particular, a key strength of the present dataset lies in the integration of anatomical markers, marker clusters, and IMUs. Because the marker clusters share the same rigid-body kinematics as the attached IMUs, they provide ground truth orientations for the corresponding IMU frames. This enables evaluation of IMU-derived orientation accuracy and analysis of how orientation accuracy affects the estimation of kinematic variables that rely on orientation information, such as linear acceleration and angular velocity expressed in the global frame. Furthermore, the effects of orientation errors on kinetic variables derived from kinematic estimates can also be investigated. Unlike overground datasets, treadmill-based protocols allow for controlled variations in speed and incline, enabling continuous gait cycles under repeatable and parametrically adjustable conditions. The dataset also incorporates functional tasks, such as squats, weight shifts, and squat jumps, providing opportunities for research beyond locomotion. Although the present dataset primarily consists of lower-body-dominant tasks, upper-body kinematics remain important even in these movements. For example, upper-body motions, such as trunk motion and arm swing during walking, can influence whole-body center of mass motion and, consequently, affect kinetic variables such as GRFs51,52. In particular, during functional movements, upper-body motions contribute to maintaining whole-body balance; thus, incorporating upper-body kinematics is essential for accurate kinetic analyses. In addition, upper-body segments are advantageous in terms of practical sensor placement and use, for example, as vest-type devices can be worn on the trunk55 and smartwatches on the wrist56. Therefore, measuring upper-body kinematics is important for understanding whole-body dynamics and for conducting studies employing wearable devices. Although this dataset is limited in participant diversity and constrained by the treadmill-based experimental protocol, it provides a useful resource for research on whole-body movement mechanics across diverse tasks. In particular, this dataset can support data-driven modeling of whole-body kinematics and kinetics, as well as inverse dynamics analyses for both gait and functional movements, thereby extending its applicability beyond conventional gait-focused studies. Methods Participants Twelve healthy young adults (10 males and 2 females, 25.2 ± 2.2 years, 78.49 ± 13.78 kg, 1.72 ± 0.06 m) participated in this study. The study protocol was approved by the Public Institutional Review Board of the Ministry of Health and Welfare of Korea (P01-202110-13-001). Participants were recruited from Hankyong National University, and informed consent was obtained from all participants prior to the experiment. The cohort was limited to healthy young adults to ensure experimental control; this restriction should be considered when generalizing the findings to older adults or patient populations. Experimental procedures Each participant performed 24 trials consisting of 2 static trials, 12 locomotion trials, and 10 functional movement trials. Two trained operators conducted all sessions to ensure data quality; one supervised the OMC, force plate, and insole recordings, whereas the other monitored the acquisition of IMU data and treadmill operation. Prior to each trial, careful checks were performed to ensure that the reflective markers, IMUs, and insoles were securely attached and remained in their initial positions. In addition, sensor placement was monitored in real time through the optical motion capture display. For the pressure-sensing insoles, positioning was checked to ensure that the heel portion of each insole was fully seated against the inside back of the shoe. A minimum rest period of 3 min (or longer, if requested by the participants) was provided between the trials. Static trials consisted of the N-pose, in which the participants stood upright with arms lowered, and the T-pose, in which the participants stood upright with arms extended horizontally to the sides. Only OMC and force plate data were collected during the static trials. Locomotion trials involved walking and running on a front-to-rear split-belt treadmill at specific speeds and inclines. Walking trials were performed at three speeds (3, 4, and 5 km/h) and three inclines (0%, 5%, and 10%), with each trial lasting for 2 min. Running trials were performed at a 0% incline and at three speeds (6, 7, and 8 km/h), with each trial lasting 90 s. The front-to-rear split-belt treadmill design enabled independent measurement of GRFs for each foot while allowing participants to walk naturally without mediolateral constraint. The participants were instructed to avoid a double stance on the same belt. Prior to the experimental trials, all participants underwent a familiarization session in which they practiced walking and running on the treadmill with visual feedback to become accustomed to the treadmill setup and to establish natural gait. Visual feedback was provided by a monitor in front of the treadmill, which displayed a real-time 3D skeleton and a lateral camera view, allowing participants to monitor their gait. Functional-movement trials consisted of in-place tasks—weight shifting (rocking side to side), half squat, full squat, side squat, and squat jump—with each task performed in two trials and with each trial lasting for 1 min. Participants were instructed to perform each movement at a self-selected pace to preserve natural movement, while allowing the dataset to capture a broader range of movement behaviors. During these trials, the participants stood facing right on a front-to-rear split-belt treadmill, with each foot positioned on a separate belt. Data acquisition for both the locomotion and functional-movement trials began with the participants standing on a step located behind the treadmill. To synchronize the treadmill and insole data, the participants struck the rear belt three times with their heels before stepping on the treadmill. This synchronization cue, combined with hardware triggers, enabled subframe alignment for all modalities. Equipment and measurement setup Lab-based systems for measuring kinematic and kinetic data comprised an OMC system and a force-sensing instrumented treadmill, whereas wearable systems consisted of an IMC system and pressure-sensing insoles. Fig. 1 shows the experimental setup and measurement scenes, and Fig. 2 shows the detailed equipment setup.Fig. 1Experimental setup and measurement scenes.Fig. 2Details of equipment setups. Experimental setup and measurement scenes. Details of equipment setups. Optical motion capture An OptiTrack Prime 13w system (NaturalPoint, Inc., Corvallis, OR, USA) with 12 cameras was used to track the trajectories of 102 reflective markers attached across the entire body. The markers comprised 45 anatomical markers placed according to anatomical landmarks, 6 additional markers for restoring lower-limb anatomical markers, and 51 markers mounted on 17 rigid IMU clusters. The anatomical landmarks were selected on the basis of a Plug-in-Gait model15. To compute the joint center locations of the upper and lower extremities more effectively, markers were placed on both the lateral and medial aspects of the elbows, wrists, knees, and ankles. Fig. 3 shows the locations of the anatomical markers. Each rigid IMU cluster was custom-fabricated using a 3D printer and included three reflective markers arranged in a triangular configuration, enabling precise estimation of the IMU cluster orientation (see Fig. 2). Depending on the body segment, the structure of the marker cluster was designed to minimize interference between markers. All the participants wore OptiTrack Velcro-type motion capture suits, which allowed for the secure attachment of markers and rigid clusters. Both anatomical markers and rigid marker clusters were mounted on the suit using Velcro-backed bases. The toe and heel markers were affixed to the shoes using adhesive tape, and the clusters were tightened with Velcro straps to ensure a secure attachment. Prior to the experiments, camera calibration and global coordinate system setup were performed. The cameras were calibrated using a CW-500 Active Calibration Wand Kit (NaturalPoint, Inc., Corvallis, OR, USA) by moving three fixed markers on the wand across the 3D capture volume. All calibrations were performed until residual mean errors were less than or equal to 1 mm to ensure reliable accuracy. The OMC global coordinate system was defined using three reference markers placed at the reference points of the treadmill (Fig. 2). Marker trajectories were recorded at a frequency of 100 Hz.Fig. 3Locations and labels of 45 anatomical markers (white: original markers; blue: additional markers attached to define joint centers). Locations and labels of 45 anatomical markers (white: original markers; blue: additional markers attached to define joint centers). Inertial motion capture Seventeen Xsens MTw 9-axis IMUs (Xsens Technologies B.V., Enschede, Netherlands) were attached to the head, sternum, shoulders, upper arms, forearms, hands, pelvis, thighs, shanks, and feet according to the sensor setup of the Xsens MVN system. Each IMU recorded triaxial accelerations (including gravitational and body accelerations), triaxial angular velocities, and triaxial magnetic field measurements (including the magnetic field of the Earth and external disturbances) in the sensor frame. Although a commercial system was employed, raw accelerometer, gyroscope, and magnetometer signals were provided to enable independent reprocessing and algorithm development beyond the vendor software. To compensate for magnetic disturbances due to the experimental environment (e.g., treadmill), magnetometer calibration was performed on the treadmill. The calibration was conducted by rotating the sensor in multiple directions according to the procedure provided in Xsens MT Manager57. The IMUs were fitted into the recesses of plastic marker clusters and secured to the body segments using Velcro straps. The marker cluster was designed such that the three reflective markers formed a plane, allowing an orthogonal coordinate system to be defined and aligned with the IMU coordinate system (see Fig. 2). Accordingly, the marker cluster data provide ground truth orientation for the corresponding IMU frame. The IMU data were wirelessly recorded at a sampling rate of 60 Hz via the Xsens Awinda Station and resampled to 100 Hz for integration with other modalities. Instrumented treadmill A front-to-rear split-belt AMTI Tandem Treadmill (AMTI, Watertown, MA, USA) with force plates embedded in each belt was used to acquire ground-reaction measurements. Each force plate measured the belt reaction forces, moments, and CoPs with respect to its local coordinate system, which were subsequently transformed into an OMC global frame and processed to compute the GRF and CoP trajectories for each foot. The origin of each belt was defined based on the position of the treadmill calibration reference point and origin offset provided by the manufacturer. For inclined walking, the global coordinate transformation of the force plate data was recalibrated for each inclination using the premeasured belt corner positions. During the locomotion trials, the participants were instructed to avoid stepping on the same belt with both feet to ensure accurate separation of the GRFs and GRMs for each foot. In the functional-movement trials, the participants stood facing right on the treadmill, with their left and right feet placed on the front and rear belts, respectively, ensuring that the force plate data from each belt corresponded to the foot placed atop it. The details of the treadmill are shown in Fig. 2. The force plate data were recorded at a sampling rate of 1,000 Hz. Pressure-sensing insoles The Novel Pedar system (Novel GmbH, Munich, Germany) was used to measure the plantar pressure in both feet. Each participant wore a waist belt equipped with an insole data logger for data transmission and storage, and each insole was connected to the data logger via a cable. The plantar pressure of each insole was measured using 99 sensor elements, and the vertical GRF and plantar CoP trajectories were computed from these 99 pressure signals. Each of the 99 insole-sensing elements was labeled sequentially from 1 to 99, as shown in the right panel of Fig. 2. Scale and bias calibrations were conducted by comparing unilateral stance data with the data obtained from the treadmill force plates to ensure the consistency of force magnitudes across systems. The insoles were inserted into the participants’ shoes without folding and positioned so that the rear part of each insole was fully seated against the inside back of the shoe. Plantar pressure data were recorded at a sampling rate of 100 Hz and stored on a data logger SD card. Synchronization device eSync2 (NaturalPoint, Inc., Corvallis, OR, USA) was used to synchronize the OMC, IMC, and instrumented treadmill data. eSync2 was connected to an OptiTrack OptiHub 2 for the OMC system, to an AMTI Optima-SC Amplifier for the force plates, and to an Xsens Awinda Station for the IMC system. At the onset of the OMC recording, a trigger signal was generated to initiate simultaneous data acquisition for the treadmill and IMC. Post-processing alignment was also performed. Angular velocity magnitudes measured using the IMU gyroscope were aligned with OMC-based cluster-derived angular velocities. Additionally, vertical GRF curves obtained from the insoles were aligned to the force- plate measurements using the data recorded during three consecutive heel strikes at the onset of each trial. This two-step approach ensured sub-frame synchronization (<0.01 s at 100 Hz) across all modalities. Data processing Marker trajectory processing, including marker labeling and gap filling, was performed using the Motive Body software (NaturalPoint, Inc., Corvallis, OR, USA). Short gaps (<0.2 s) were filled using cubic interpolation, whereas longer gaps (0.2 s–1.0 s) were reconstructed using pattern-based interpolation. Pattern-based interpolation was performed using three or more reference markers located on the same segment to reconstruct missing markers based on the relationships among those markers, and was applied using the pattern interpolation function provided in Motive Body software58. Gaps longer than 1.0 s were considered difficult to reconstruct accurately; in such cases, the trial was repeated. When marker swaps or jitter occurred due to occlusion, trajectories were visually inspected using lateral camera recordings and corrected manually before reapplying the interpolation. These quality control steps ensured the integrity of the biomechanical model. Processed marker trajectories and force plate data were exported in C3D format. To compute full-body kinematic and kinetic data, they were processed in Visual3D (C-Motion, Inc., Germantown, MD, USA) which has been widely used for biomechanical modeling and inverse dynamics analysis in the field of motion analysis59–61. The global coordinate system was defined with the X-axis oriented along the direction of treadmill progression, the Y-axis oriented leftward, and the Z-axis oriented vertically. Walking data obtained from the treadmill at various inclinations were mapped onto a common frame using the positions of belt-corner markers premeasured at each incline. For biomechanical modeling, subject-specific parameters (body mass, height, and hand thickness) were used as inputs. Subsequently, the coordinate frames, geometries, inertial parameters, and associated landmarks were defined using the marker trajectories in the static trials. Segment coordinate frames were defined using vectors pointing from distal to proximal landmarks (Z-axis) and those oriented along the medial-to-lateral direction (Y-axis, reversed for the right limbs). Inertial parameters were computed using established anthropometric models within Visual3D: segment mass was estimated based on Dempster62, while the center of mass location and moment of inertia were defined based on segment geometries derived according to Hanavan63. For the segment geometries, the thorax and pelvis were modelled as cylinders; the head and hands as ellipsoids; and the upper and lower limb segments as truncated cones. Using these definitions, a biomechanical model was generated from the static trials, enabling the computation of joint centers, segment orientations, and kinetic data via inverse dynamics. It should be noted that, in the present dataset, the biomechanical modeling procedure was based solely on the OMC data, whereas the IMU data were independent of the anatomical model, as they consisted only of sensor signals and IMU frame orientations. An overview of the segment frame definitions and biomechanical model is shown in Fig. 4. The resulting processed outputs, including marker trajectories, force plate data, and biomechanical variables computed using the biomechanical model, were exported in the MAT format. All the Visual3D processing steps were automated using a custom pipeline to ensure consistency and reproducibility.Fig. 4Illustration of the segment-frame definition and biomechanical model on Visual3D and MATLAB. Illustration of the segment-frame definition and biomechanical model on Visual3D and MATLAB. The IMC data were extracted as TXT files using the Xsens MT Manager, whereas the pressure-sensing insole data were extracted using Novel Pedar software. Pressure data from 198 sensing elements were stored as ASC files, and vertical GRF and plantar CoP trajectories were stored as MVA files. MATLAB (MathWorks, Inc., Natick, MA, USA) was used to integrate and process the data extracted from Visual3D, along with the IMC and insole data. The positions and orientations of the IMU reference frames within the OMC were determined based on the marker cluster trajectories. Subsequently, the 9-axis IMU signals originally sampled at 60 Hz were resampled to 100 Hz using cubic spline interpolation to match the sampling frequency of the other modalities (i.e., OMC, force plates, and pressure-sensing insoles), as in prior studies integrating OMC and IMC data64,65. Given the temporal characteristics of the target movements in relation to the 60 Hz sampling rate, the resampling process was unlikely to introduce substantial signal distortion, and visual inspection confirmed that the resampled signals preserved the overall profiles of the original data. Further time synchronization between the IMC and OMC systems was performed by aligning the angular velocity magnitudes from the resampled gyroscope signals with those derived from the marker cluster orientations, thereby correcting minor temporal discrepancies between the two modalities. When IMU data loss occurred due to wireless transmission failure, virtual IMU signals were generated from marker cluster data to bridge the gaps by interpolation, given that the marker clusters and IMUs share the same rigid-body kinematics. Time synchronization between the insole and force plate data was achieved by aligning the vertical GRF curves measured by the two systems during three consecutive heel-strike impacts on the rear belt at the beginning of each trial. For the coordinate mapping of the insole data, the offset between the heel marker and insole was determined using the plantar CoP trajectories obtained from the projected force plate data and the insoles in each foot plane. The data from the four systems integrated at a sampling rate of 100 Hz in MATLAB were exported as MAT files. Each MAT file included structured arrays for markers, IMUs, force plates, insoles, and biomechanical segment variables, ensuring transparency and compatibility with both MATLAB and Python environments. A flowchart of the pipeline is shown in Fig. 5.Fig. 5Flowchart of data processing pipeline. Flowchart of data processing pipeline. Data Records All dataset files are available in the online Zenodo repository54. The dataset was distributed in two primary formats: C3D files containing OMC and force plate data and MAT files containing synchronized multimodal sensor data and the derived biomechanical variables. All data were organized into subject-specific folders named by subject identifier (formatted as S01, S02, and so on), each containing all files for a single subject. The C3D and MAT files are stored in separate directories to distinguish the raw optical data from the processed multimodal dataset. All C3D and MAT file names were formatted as _ (e.g., “S01_Npose”), where all 24 trial names are summarized in Table 2. In addition, we used an Excel file containing the experimental trial information, marker lists, segment information, variable lists, force plate corner locations, insole-sensing areas, insole factors, and subject information.Table 2Names of all 24 trials.No.Trial typeTrial nameSpeedIncline1Static trialNposeNA02Static trialTposeNA03Locomotion trialWalking_3KPH_0304Walking_4KPH_0405Walking_5KPH_0506Walking_3KPH_5357Walking_4KPH_5458Walking_5KPH_5559Walking_3KPH_1031010Walking_4KPH_1041011Walking_5KPH_1051012Running_6KPH6013Running_7KPH7014Running_8KPH8015Functional movement trialWeight_Shift_1NA016Half_Squat_1NA017Full_Squat_1NA018Side_Squat_1NA019Squat_Jump_1NA020Weight_Shift_2NA021Half_Squat_2NA022Full_Squat_2NA023Side_Squat_2NA024Squat_Jump_2NA0 Table 2Names of all 24 trials.No.Trial typeTrial nameSpeedIncline1Static trialNposeNA02Static trialTposeNA03Locomotion trialWalking_3KPH_0304Walking_4KPH_0405Walking_5KPH_0506Walking_3KPH_5357Walking_4KPH_5458Walking_5KPH_5559Walking_3KPH_1031010Walking_4KPH_1041011Walking_5KPH_1051012Running_6KPH6013Running_7KPH7014Running_8KPH8015Functional movement trialWeight_Shift_1NA016Half_Squat_1NA017Full_Squat_1NA018Side_Squat_1NA019Squat_Jump_1NA020Weight_Shift_2NA021Half_Squat_2NA022Full_Squat_2NA023Side_Squat_2NA024Squat_Jump_2NA0 Names of all 24 trials. The C3D files contained 102 marker trajectories, GRFs, GRMs, and CoP trajectories for each belt of the front-to-rear belt treadmill. They also included the associated metadata, such as information on the OMC system, corner positions, and origin offsets of the force plates for the front and rear belts, as well as software and participant information. The MAT files contained the following structures: Marker, which stored anatomical marker trajectories; IMU, which stored IMU signals and marker cluster data for 17 sensor attachment locations; FP, which stored data obtained from the force plates; Insole, which stored the data obtained from the pressure-sensing insoles; Landmark, which stored joint center positions; Segment, which stored the kinematic and kinetic data computed from 15 body segments using the biomechanical model. In the data description, the number of samples was denoted by N, an array with A rows and B columns was denoted by (A,B), and the fields or variables of the structure were denoted by . In all cases, the 3D vectors were organized in the order of X–Y–Z, and the rotation matrices were represented as 3 × 3 matrices, with each column corresponding to the X-, Y-, and Z-axes of the local frame expressed in the global frame. The Marker structure contained the trajectories of 45 anatomical markers, each stored in a (3,N) array in the format Marker. , where the labels of are listed in Table 3.Table 3List of marker labels, related segments, and marker placements (see Table 4 for segment label definition).No.LabelRelated segmentPlacement description1RFHDHERight side of the anterior head2RBHDHERight side of the posterior head3LFHDHELeft side of the anterior head4LBHDHELeft side of the posterior head5CLAVTACenter of the two clavicle bones6STRNTASternum xiphoid process7C7TA7th cervical spine vertebra8T10TA10th thoracic spine vertebra9RBAKTARight posterior-lateral side of trunk10RSHOTA,RUARight acromial prominence11RUPARUALateral side of right upper arm12RELBLRUA,RFALateral epicondyle of right elbow13RELBMRUA,RFAMedial epicondyle of right elbow14RFRMRFAUlnar side of right forearm15RWRARFA,RHARight ulna styloid process16RWRBRFA,RHARight radius styloid process17RFINRHA2nd metatarsal of right hand18LSHOTA,LUALeft acromial prominence19LUPALUALateral side of left upper arm20LELBLLUA,LFALateral epicondyle of left elbow21LELBMLUA,LFAMedial epicondyle of left elbow22LFRMLFAUlnar side of left forearm23LWRALFA,RHALeft ulna styloid process24LWRBLFA,RHALeft radius styloid process25LFINLHA2nd metatarsal of left hand26RASIPVRight anterior superior iliac spine27RPSIPVRight posterior superior iliac spine28LASIPVLeft anterior superior iliac spine29LPSIPVLeft posterior superior iliac spine30RTHIRTHLateral side of right thigh31RKNELRTH,RSKLateral epicondyle of right knee32RKNEMRTH,RSKMedial epicondyle of right knee33RTIBRSKLateral side of right tibia34RANKLRSK,RFTLateral malleolus of right ankle35RANKMRSK,RFTMedial malleolus of right ankle36RTOERFT2nd metatarsal of right foot37RHEERFTCalcaneus of right foot38LTHILTHLateral side of left thigh39LKNELLTH,LSKLateral epicondyle of left knee40LKNEMLTH,LSKMedial epicondyle of left knee41LTIBLSKLateral side of left tibia42LANKLLSK,LFTLateral malleolus of left ankle43LANKMLSK,LFTMedial malleolus of left ankle44LTOELFT2nd metatarsal of left foot45LHEELFTCalcaneus of left foot Table 3List of marker labels, related segments, and marker placements (see Table 4 for segment label definition).No.LabelRelated segmentPlacement description1RFHDHERight side of the anterior head2RBHDHERight side of the posterior head3LFHDHELeft side of the anterior head4LBHDHELeft side of the posterior head5CLAVTACenter of the two clavicle bones6STRNTASternum xiphoid process7C7TA7th cervical spine vertebra8T10TA10th thoracic spine vertebra9RBAKTARight posterior-lateral side of trunk10RSHOTA,RUARight acromial prominence11RUPARUALateral side of right upper arm12RELBLRUA,RFALateral epicondyle of right elbow13RELBMRUA,RFAMedial epicondyle of right elbow14RFRMRFAUlnar side of right forearm15RWRARFA,RHARight ulna styloid process16RWRBRFA,RHARight radius styloid process17RFINRHA2nd metatarsal of right hand18LSHOTA,LUALeft acromial prominence19LUPALUALateral side of left upper arm20LELBLLUA,LFALateral epicondyle of left elbow21LELBMLUA,LFAMedial epicondyle of left elbow22LFRMLFAUlnar side of left forearm23LWRALFA,RHALeft ulna styloid process24LWRBLFA,RHALeft radius styloid process25LFINLHA2nd metatarsal of left hand26RASIPVRight anterior superior iliac spine27RPSIPVRight posterior superior iliac spine28LASIPVLeft anterior superior iliac spine29LPSIPVLeft posterior superior iliac spine30RTHIRTHLateral side of right thigh31RKNELRTH,RSKLateral epicondyle of right knee32RKNEMRTH,RSKMedial epicondyle of right knee33RTIBRSKLateral side of right tibia34RANKLRSK,RFTLateral malleolus of right ankle35RANKMRSK,RFTMedial malleolus of right ankle36RTOERFT2nd metatarsal of right foot37RHEERFTCalcaneus of right foot38LTHILTHLateral side of left thigh39LKNELLTH,LSKLateral epicondyle of left knee40LKNEMLTH,LSKMedial epicondyle of left knee41LTIBLSKLateral side of left tibia42LANKLLSK,LFTLateral malleolus of left ankle43LANKMLSK,LFTMedial malleolus of left ankle44LTOELFT2nd metatarsal of left foot45LHEELFTCalcaneus of left foot List of marker labels, related segments, and marker placements (see Table 4 for segment label definition). The IMU structure contained the trajectories of the three cluster markers, positions, and orientation matrices of the IMU frame derived from the marker clusters, 3-axis accelerometer signals, 3-axis gyroscope signals, 3-axis magnetometer signals, and orientation matrices, estimated by the onboard Kalman filter, of the 17 IMUs. Among these, the orientation matrices were stored as (3,3,N) arrays, whereas all other vectors were stored as (3,N) arrays in the format IMU. ., where the definitions of and are listed in Tables 4, 5, respectively.Table 4List of 17 IMU attachment locations and the corresponding segments with proximal joint.No.LabelSegment/LocationProximal joint1HEHeadNeck2TAThoraxLumbar3PVPelvisLumbar4RUARight upper armRight shoulder5RFARight forearmRight elbow6RHARight handRight wrist7RTHRight thighRight hip8RSKRight shankRight knee9RFTRight footRight ankle10LUALeft upper armLeft shoulder11LFALeft forearmLeft elbow12LHALeft handLeft wrist13LTHLeft thighLeft hip14LSKLeft shankLeft knee15LFTLeft footLeft ankle16RSORight shoulderNA17LSOLeft shoulderNATable 5List of variables in the IMU structure.No.Variable nameDescriptionUnitDimension1Marker11st cluster marker positionm(3,N)2Marker22nd cluster marker positionm(3,N)3Marker33rd cluster marker positionm(3,N)4IMURotIMU orientation matrixNA(3,3,N)5IMUPosIMU positionm(3,N)6AccAccelerometer signalm/s2(3,N)7GyroGyroscope signalrad/s(3,N)8MagMagnetometer signala.u.(3,N)9IMURotKFIMU orientation matrix from onboard Kalman filterNA(3,3,N) Table 4List of 17 IMU attachment locations and the corresponding segments with proximal joint.No.LabelSegment/LocationProximal joint1HEHeadNeck2TAThoraxLumbar3PVPelvisLumbar4RUARight upper armRight shoulder5RFARight forearmRight elbow6RHARight handRight wrist7RTHRight thighRight hip8RSKRight shankRight knee9RFTRight footRight ankle10LUALeft upper armLeft shoulder11LFALeft forearmLeft elbow12LHALeft handLeft wrist13LTHLeft thighLeft hip14LSKLeft shankLeft knee15LFTLeft footLeft ankle16RSORight shoulderNA17LSOLeft shoulderNA List of 17 IMU attachment locations and the corresponding segments with proximal joint. Table 5List of variables in the IMU structure.No.Variable nameDescriptionUnitDimension1Marker11st cluster marker positionm(3,N)2Marker22nd cluster marker positionm(3,N)3Marker33rd cluster marker positionm(3,N)4IMURotIMU orientation matrixNA(3,3,N)5IMUPosIMU positionm(3,N)6AccAccelerometer signalm/s2(3,N)7GyroGyroscope signalrad/s(3,N)8MagMagnetometer signala.u.(3,N)9IMURotKFIMU orientation matrix from onboard Kalman filterNA(3,3,N) List of variables in the IMU structure. The FP structure contained the measured GRFs, GRMs, and CoPs from the front and rear belts, the calculated GRFs, GRMs, and CoPs for the left and right feet, and the corner positions and origin of the force plates for each belt. The GRFs, GRMs, and CoPs were stored as (3,N) arrays, corner positions as (3,4) arrays, and origins as (3,1) arrays in the format FP. , where the labels of are listed in Table 6.Table 6List of variables in the FP structure (GRF: ground reaction force, GRM: ground reaction moment, CoP: center of pressure).No.Variable nameDescriptionUnitDimension1FBGRFFront belt’s GRFN(3,N)2FBGRMFront belt’s GRMNm(3,N)3FBCoPFront belt’s CoPm(3,N)4RBGRFRear belt’s GRFN(3,N)5RBGRMRear belt’s GRMNm(3,N)6RBCoPRear belt’s CoPm(3,N)7RGRFRight foot’s GRFN(3,N)8RCoPRight foot’s CoPm(3,N)9LGRFLeft foot’s GRFN(3,N)10LCoPLeft foot’s CoPm(3,N)11FBCornerFront belt’s corner positionsm(3,4)12FBOriginFront belt’s origin positionm(3,1)13RBCornerRear belt’s corner positionsm(3,4)14RBOriginRear belt’s origin positionm(3,1) Table 6List of variables in the FP structure (GRF: ground reaction force, GRM: ground reaction moment, CoP: center of pressure).No.Variable nameDescriptionUnitDimension1FBGRFFront belt’s GRFN(3,N)2FBGRMFront belt’s GRMNm(3,N)3FBCoPFront belt’s CoPm(3,N)4RBGRFRear belt’s GRFN(3,N)5RBGRMRear belt’s GRMNm(3,N)6RBCoPRear belt’s CoPm(3,N)7RGRFRight foot’s GRFN(3,N)8RCoPRight foot’s CoPm(3,N)9LGRFLeft foot’s GRFN(3,N)10LCoPLeft foot’s CoPm(3,N)11FBCornerFront belt’s corner positionsm(3,4)12FBOriginFront belt’s origin positionm(3,1)13RBCornerRear belt’s corner positionsm(3,4)14RBOriginRear belt’s origin positionm(3,1) List of variables in the FP structure (GRF: ground reaction force, GRM: ground reaction moment, CoP: center of pressure). The Insole structure contained the pressure signals and sensing areas of 99 sensing elements for each insole, the vertical GRFs and horizontal CoPs computed from the pressure signals, and the offset between the heel marker and insole origin. The pressure signals were stored as (N,99) arrays, sensing areas as (1,99) arrays, vertical GRFs as (N,1) arrays, horizontal CoPs as (N,2) arrays, and heel–insole offsets as (1,2) arrays in the format Insole. , where the definitions of are listed in Table 7. In the CoP trajectory and offset arrays, the first and second columns correspond to frontal and lateral components, respectively.Table 7List of variables in the Insole structure (GRF: ground reaction force, CoP: center of pressure).No.Variable nameDescriptionUnitDimension1RGRFRight insole’s vertical GRFN(N,1)2RCoPRight insole’s plantar CoPm(N,2)3RPressureRight insole’s pressure signalN/mm2(N,99)4LGRFLeft insole’s vertical GRFN(N,1)5LCoPLeft insole’s plantar CoPm(N,2)6LPressureLeft insole’s pressure signalN/mm2(N,99)7AreaSensing element areamm2(1,99)8ROrigOffsetRight heel-insole offsetm(1,2)9LOrigOffsetLeft heel- insole offsetm(1,2) Table 7List of variables in the Insole structure (GRF: ground reaction force, CoP: center of pressure).No.Variable nameDescriptionUnitDimension1RGRFRight insole’s vertical GRFN(N,1)2RCoPRight insole’s plantar CoPm(N,2)3RPressureRight insole’s pressure signalN/mm2(N,99)4LGRFLeft insole’s vertical GRFN(N,1)5LCoPLeft insole’s plantar CoPm(N,2)6LPressureLeft insole’s pressure signalN/mm2(N,99)7AreaSensing element areamm2(1,99)8ROrigOffsetRight heel-insole offsetm(1,2)9LOrigOffsetLeft heel- insole offsetm(1,2) List of variables in the Insole structure (GRF: ground reaction force, CoP: center of pressure). The Segment structure contained the segment orientation quaternion, angular velocity, proximal end position, center of mass position, and force and torque at the proximal point (i.e., the joint reaction force and torque), which were calculated using the biomechanical model for the 15 body segments. The quaternions were stored as (4,N) arrays, whereas all others were stored as (3,N) arrays in the format Segment. ., where the definitions of and are listed in Tables 4, 8, respectively.Table 8List of variables in the Segment structure (CoM: center of mass).No.Variable nameDescriptionUnitDimension1SegRotSegment orientation matrixNA(3,3,N)2AngVelSegment angular velocityrad/s(3,N)3AngAccSegment angular accelerationrad/s2(3,N)4CoMPosSegment CoM positionm(3,N)5CoMVelSegment CoM velocitym/s(3,N)6CoMAccSegment CoM accelerationm/s2(3,N)7ProxEndPosSegment proximal end positionm(3,N)8DistEndPosSegment distal end positionm(3,N)9ProxEndForceSegment proximal end forceN(3,N)10ProxEndTorqueSegment proximal end torqueNm(3,N) Table 8List of variables in the Segment structure (CoM: center of mass).No.Variable nameDescriptionUnitDimension1SegRotSegment orientation matrixNA(3,3,N)2AngVelSegment angular velocityrad/s(3,N)3AngAccSegment angular accelerationrad/s2(3,N)4CoMPosSegment CoM positionm(3,N)5CoMVelSegment CoM velocitym/s(3,N)6CoMAccSegment CoM accelerationm/s2(3,N)7ProxEndPosSegment proximal end positionm(3,N)8DistEndPosSegment distal end positionm(3,N)9ProxEndForceSegment proximal end forceN(3,N)10ProxEndTorqueSegment proximal end torqueNm(3,N) List of variables in the Segment structure (CoM: center of mass). The Joint structure contained joint angles, joint reaction forces, and torques for the shoulders, elbows, wrists, hips, knees, and ankles. Joint angles follow the Y–Z–X Euler sequence for the ankles and the Y–X–Z Euler sequence for all other joints. Joint forces and torques were expressed in the proximal segment coordinate system. Joint angles were stored as (N,3) arrays in the order of (X,Y,Z), whereas joint forces and torques were stored as (3,N) arrays under the format Joint. ., with the definitions of and provided in Tables 9, 10, respectively.Table 9List of joint locations and the corresponding proximal/distal segments.No.LabelJointProximal segmentDistal segment1RShoulderRight shoulderTARUA2RElbowRight elbowRUARFA3RWristRight wristRFARHA4RHipRight hipPVRTH5RKneeRight kneeRTHRSK6RAnkleRight ankleRSKRFT7LShoulderLeft shoulderTALUA8LElbowLeft elbowLUALFA9LWristLeft wristLFALHA10LHipLeft hipPVLTH11LKneeLeft kneeLTHLSK12LAnkleLeft ankleLSKLFTTable 10List of variables in the Joint structure.No.Variable nameDescriptionUnitDimension1AngleJoint angledeg(N,3)2ForceJoint reaction forceN(3,N)3TorqueJoint torqueNm(3,N) Table 9List of joint locations and the corresponding proximal/distal segments.No.LabelJointProximal segmentDistal segment1RShoulderRight shoulderTARUA2RElbowRight elbowRUARFA3RWristRight wristRFARHA4RHipRight hipPVRTH5RKneeRight kneeRTHRSK6RAnkleRight ankleRSKRFT7LShoulderLeft shoulderTALUA8LElbowLeft elbowLUALFA9LWristLeft wristLFALHA10LHipLeft hipPVLTH11LKneeLeft kneeLTHLSK12LAnkleLeft ankleLSKLFT List of joint locations and the corresponding proximal/distal segments. Table 10List of variables in the Joint structure.No.Variable nameDescriptionUnitDimension1AngleJoint angledeg(N,3)2ForceJoint reaction forceN(3,N)3TorqueJoint torqueNm(3,N) List of variables in the Joint structure. Technical Validation Calibration of equipment To ensure reliable measurement accuracy, all the systems were calibrated according to the manufacturer’s guidelines before each session. The OMC cameras were calibrated by moving a three-marker wand across the capture volume, with residual mean errors being consistently less than or equal to 1 mm across all sessions. Zero-offset calibration of the force plates was performed before each trial and before the participants stepped onto the treadmill. Magnetometer calibration was performed using dedicated software to compensate for time-invariant magnetic disturbances and sensor biases such that the calibrated magnetometer signals formed a sphere centered at the origin. Zero-offset calibration of the insoles was conducted under software guidance, during which the participants stood sequentially on each foot to measure the zero offsets. Additionally, scale and bias calibrations were validated by comparing unilateral stance forces with those obtained from the treadmill force plates, with mean discrepancies below 5%. Time-synchronization As the four systems operated with different data acquisition devices and sampling rates, time synchronization and result validation were performed to ensure reliable data integration. Although the OMC and IMC systems were synchronized using trigger signals from the synchronization device, a slight temporal offset occurred owing to the lower sampling rate of the IMC. To eliminate this temporal offset, the IMU-based and marker cluster-based angular velocity curves were temporally aligned until the two signals were consistent with each other. To refine the alignment, the magnitudes of angular velocity derived using the gyroscope were compared to those derived using the marker clusters after 8-Hz low-pass filtering, and the curves were temporally shifted until maximal consistency was achieved. This post-processing step reduced residual offsets to below 0.01 s (one frame at 100 Hz). The right shank IMU was used as the synchronization reference for the locomotion trials, and the right thigh IMU was used for the functional-movement trials to ensure a clearer comparison of angular-velocity curves. Fig. 6 illustrates an example of the comparison of angular-velocity curves obtained using the gyroscopes for each of the 17 IMUs and the corresponding marker cluster after the synchronization of the walking trials. The data obtained from the force plates and pressure-sensing insoles were synchronized by matching the vertical GRF curves obtained during three consecutive heel strikes recorded at the beginning of each trial. Fig. 7 illustrates a comparison between the vertical GRF curves of both feet during walking, as measured using the force plates and pressure-sensing insoles after time synchronization. During the time-synchronization process, all trials were visually inspected to confirm that the signals from both system pairs, IMU-OMC and insole-force plate, were temporally well aligned, as illustrated in Figs. 6, 7.Fig. 6Comparison of angular-velocity curves measured using marker clusters and gyroscopes for 17 IMUs [unit: rad/s].Fig. 7Comparison of vertical ground reaction force curves obtained using force plates and insole pressure sensors [unit: N]. Comparison of angular-velocity curves measured using marker clusters and gyroscopes for 17 IMUs [unit: rad/s]. Comparison of vertical ground reaction force curves obtained using force plates and insole pressure sensors [unit: N]. To verify consistency across systems, vertical GRFs from the insoles were compared with the measurements obtained from the treadmill force plates. The temporal profiles were highly correlated (average r = 0.988), with the highest correlation observed for weight shift (r = 0.997) and the lowest for half squat (r = 0.906) (see Fig. 7), supporting the validity of the insole calibration and synchronization procedures. Similarly, the angular velocities measured by the IMUs closely matched those derived from the OMC marker clusters after temporal alignment (average r = 0.936), with the highest correlation observed for walking (r = 0.957) and the lowest for side squat (r = 0.840) (see Fig. 6). It should be noted, however, that differences between the wearable sensor signals and the laboratory-based reference signals are expected and do not necessarily indicate poor signal quality in either modality. IMUs directly measure angular velocity using gyroscopes, whereas OMC-based angular velocity is obtained through an additional computational process in which sensor orientation is first estimated from the measured marker trajectories and angular velocity is then derived from that orientation. Likewise, force plates directly measure 3D ground reaction forces using fixed force sensors, whereas pressure-sensing insoles estimate vertical ground reaction force from the pressure distribution at the foot-shoe interface. Given these differences, the observed discrepancies are more likely to reflect differences in measurement principles and sensor characteristics than deficiencies in signal quality. Signal reliability Marker trajectory occlusions were corrected using the interpolation methods described in the Data Processing section. Wireless transmission failures occasionally cause brief IMU data dropouts. These gaps were filled with virtual IMU signals reconstructed from the OMC cluster orientations, ensuring the continuity of the full-body IMU streams. The IMU and marker cluster data can be regarded as measurements of the same underlying rigid-body motion obtained through different sensing modalities. In addition, previous studies have demonstrated the feasibility of simulating IMU signals from OMC data41,66. Accordingly, the reconstructed signals were considered a reasonable approximation of the missing IMU data. Regarding the extent of data loss, only one exceptional trial was identified, in which 9 samples were missing from a single sensor (corresponding to approximately 0.2% of the total samples), and all other instances were limited to very short gaps of no more than 3 samples. Therefore, the vast majority of the dataset consists of the original IMU signals. For completeness, the number of missing samples for the 17 IMUs in each trial were organized into an Excel summary file and uploaded to the dataset repository. Usage Notes C3D files The dataset was prepared in formats that are widely accessible to the biomechanics and data science communities. The C3D files contain optical marker trajectories and treadmill force plate data, and they are compatible with commercial software such as Visual3D, Vicon Nexus, and Qualisys Track Manager, as well as open-source platforms, including OpenSim. Using libraries such as Biomechanical ToolKit (BTK)67, C3D files can also be imported into MATLAB or Python environments. In addition, the open-source software Mokka68 included in the BTK suite allows users to visualize, inspect, and manage C3D files through a graphical user interface. MAT files The MAT files containing raw measurements and variables derived from the four systems were accessible and processable using both MATLAB and Python. However, MATLAB is recommended for data exploration and preprocessing, as it enables an intuitive inspection of fields within structural variables (e.g., via the Workspace/Variable Editor). MATLAB also supports libraries for IMU-based orientation estimation (e.g., AHRS in the sensor fusion and Tracking Toolbox69), handling orientation representations (e.g., rotation matrix-to-Euler conversions via the Robotics System Toolbox70), and signal filtering and visualization (e.g., Signal Processing Toolbox71). Potential applications The dataset is suitable for a wide range of applications, including the validation of wearable-based algorithms against laboratory gold standards and training of machine learning models. For example, the dataset can be used for predicting GRFs or gait phases from IMU signals and for inverse dynamics studies using ready-to-use joint torque and reaction force data. It can also be applied for comparative analysis across locomotion conditions, including speed- and incline-dependent adaptations. Limitations and future directions The main limitation of this dataset is that only 12 healthy young participants were included, predominantly male (10 males and 2 females), which limits its ability to fully represent diverse age groups, clinical populations, and variations in kinematic and kinetic characteristics between male and female participants. Considering that gender-related differences in gait kinematics and kinetics have been reported72, this gender imbalance not only hinders the adequate reflection of biomechanical differences between males and females but also restricts the representation of this age group’s characteristics. Another limitation is that all locomotion trials were conducted exclusively on a treadmill, providing controlled and repeatable conditions. Although treadmill-based experiments enable prolonged walking under predefined settings (e.g., speed and incline), they do not capture the variability of overground gait or daily life movements encountered in real-world applications. For example, although treadmill and overground walking exhibit similar patterns in terms of kinematics, differences have been reported in joint moments and ground reaction forces73. Therefore, when applying models or analyses derived from this dataset to real-world scenarios, differences in the resulting outcomes may arise and should be carefully considered. To address the limitations in participant sample composition, future studies should recruit a more balanced gender distribution and a larger number of participants, and may further expand the study population to include diverse groups such as older adults and clinical populations. Furthermore, to improve applicability in both real-world and sport-related scenarios, future studies should incorporate a wider range of movement tasks, including overground gait, upper-body functional tasks, and sport-related movements.