Study at a glance

In nine healthy adults adapting to a powered ankle exoskeleton during treadmill walking, Theia3D markerless motion capture tracked how each individual's lower-limb kinematics evolved over time, revealing that motor adaptation was gradual, highly individualized, and occurred asynchronously across the ankle, knee, and hip, with group-averaged analysis obscuring much of this inter-participant variability.

Summary

Disclaimer: The following summary details independent academic research. Theia3D is intended for research, sports performance and educational use only. It is not cleared as a medical device and is not intended to be used for the diagnosis, treatment, mitigation or prevention of any disease, injury or medical condition.

Editor's note: Theia3D version not specified in the paper. This study uses Theia3D as a measurement tool to track motor adaptation during exoskeleton-assisted walking, rather than evaluating Theia3D's own accuracy; the paper cites prior Theia3D validation literature for lower-limb sagittal-plane kinematics as the basis for using it in this application.

Study Overview

Design

A within-subject study tracking the continuous evolution of lower-limb kinematics, inter-joint coordination, and metabolic cost of transport as participants adapted to treadmill walking with a powered ankle exoskeleton, compared with walking without the exoskeleton and with the exoskeleton worn but providing no torque assistance.

Participants

Nine healthy young adults (6 women, 3 men; mean age 25.7 years; mean height 171.6 cm; mean weight 69.0 kg) with no ongoing history of lower-limb musculoskeletal or mobility-related conditions, and no prior experience with the specific exoskeleton used in this study.

Data Collection

Each participant completed three treadmill walking conditions at their own preferred walking speed: 6 minutes without the exoskeleton (baseline), 25 minutes with the exoskeleton providing bidirectional torque assistance at the ankle during both plantarflexion and dorsiflexion, and 6 minutes with the exoskeleton worn but set to provide zero torque. Eight synchronized Sony RX0 II cameras recorded lower-limb kinematics throughout, processed through Theia3D to generate continuous ankle, knee, and hip joint angles in the sagittal plane, while a portable indirect calorimetry system recorded metabolic cost of transport throughout the exoskeleton-assisted condition.

Analysis

Lower-limb kinematic trajectories were segmented into stance and swing phases using automated, kinematics-based gait event detection, then grouped into 15-second time windows to track how each joint's trajectory evolved and converged over the course of each condition, using inter-trajectory root-mean-square error against each participant's own stabilized end-of-condition trajectory. Inter-joint coordination was separately assessed using three-dimensional cyclograms of the ankle, knee, and hip elevation angles, with a linear dependency metric tracking how consistently these joints moved together over time. Metabolic cost of transport was calculated relative to each participant's own baseline walking condition, and participants separately self-reported, via a simple hand-raise cue, when they subjectively felt they had adapted to the exoskeleton.

Key Findings

  • Motor adaptation to exoskeleton-assisted walking was gradual and highly individualized, with substantial variability in both the timing and magnitude of kinematic convergence across participants. While most participants eventually achieved kinematic convergence at the ankle and hip, all but one participant failed to reach the study's convergence threshold for knee trajectories within the 25-minute exposure.
  • Kinematic adaptation occurred asynchronously across the lower limb, rather than all joints settling into a stable pattern at the same time. The ankle, where exoskeleton assistance was applied directly, frequently showed the largest initial deviations from baseline walking, while convergence timing and the specific joint sequence varied considerably from participant to participant.
  • Inter-trajectory variability was consistently larger during the swing phase of gait than during stance, a pattern observed across all participants. The authors attribute this to the exoskeleton and leg effectively operating as an open kinetic chain during swing, with no ground contact to help constrain ankle motion, compared with the more constrained, closed kinetic chain of the stance phase.
  • Metabolic cost of transport did not decrease overall with exoskeleton assistance, and responses were highly heterogeneous across participants. On average, participants showed a 5% increase in metabolic cost of transport relative to walking without the exoskeleton, though five of seven participants with usable metabolic data did show a decreasing trend over the course of the exoskeleton-assisted condition.
  • Group-averaged kinematic analysis substantially understated the individual variability present in the data, with inter-trajectory error in the group-averaged results converging almost immediately and appearing far more uniform than the considerable differences observed when each participant's data was examined individually, a pattern the authors argue could lead to overestimating how consistently and quickly exoskeleton users actually adapt if only group-level data were considered.

What This Means for Movement Science Researchers and Exoskeleton Development Teams

For teams developing or evaluating powered exoskeletons for occupational, defence, or mobility-support applications, this study's central message is methodological as much as it is a specific finding: individual-level, time-resolved kinematic analysis reveals meaningful adaptation patterns, including joints that never fully stabilize and substantial variation in adaptation timing, that a group-averaged, steady-state comparison would miss entirely. The authors' finding that self-reported adaptation often didn't align with objective kinematic or metabolic convergence is also a useful caution against relying on subjective adaptation reports alone when evaluating how well a given user has adjusted to an assistive device. More broadly, this study is a practical demonstration of how markerless motion capture's ability to track continuous movement without marker occlusion, even through the awkward process of donning and removing a bulky exoskeleton, supports exactly the kind of detailed, longitudinal, individual-level analysis the authors argue the field needs more of.

The Role of Theia3D in Movement Analysis

Theia3D served as the kinematic measurement tool throughout this study rather than as a system under direct evaluation, continuously tracking ankle, knee, and hip joint angles across three walking conditions as participants wore, adapted to, and transitioned out of a powered ankle exoskeleton. The authors note that markerless capture allowed participants' kinematic data to be generated without occlusion even while the exoskeleton was donned on the body, a practical advantage over marker-based alternatives for this kind of continuous, device-based adaptation study, and cited prior Theia3D validation literature on lower-limb sagittal-plane accuracy as the basis for its use here.

Study Limitations and Future Directions

The authors note that participants walked at a fixed treadmill speed throughout the study, which constrains the spatiotemporal variability seen in self-paced or overground walking and likely limited the range of temporal gait adaptations observed; they suggest a self-paced or overground walking condition might reveal more pronounced adaptation in stride timing and cadence. Two participants were unable to complete the Exo Zero-Torque task due to technical limitations of the exoskeleton, and indirect calorimetry data for two other participants were affected by premature sensor failure, reducing the usable sample for some analyses. The 25-minute exposure window used in this study likely captures only an early, transient phase of motor adaptation rather than a full steady-state adaptation process, since prior literature suggests complete convergence may require extended or multi-session exposure. The choice of a 15-second window for discretizing kinematic data involved a tradeoff between noise suppression and preserving rapid adaptation dynamics, and the authors note that shorter windows made trend identification less reliable, particularly during the more variable exoskeleton-assisted condition. All participants in this study were healthy young adults, so these findings on adaptation strategy may not generalize to populations with more diverse neuromuscular profiles, who the authors suggest may show an even broader range of individualized adaptation strategies. The authors identify incorporating psychological or familiarization-related metrics alongside biomechanical and metabolic measures, and extending this individualized analysis framework to larger and more diverse cohorts, as clear next steps.

Citation: Lee, P. S., & Mombaur, K. (2026). Identifying the key biomechanical features of movement adaptation during exoskeleton-assisted locomotion. IEEE Transactions on Medical Robotics and Bionics (preprint). https://doi.org/10.48550/arXiv.2608.07140

Link: Read the full preprint

What Does Individual-Level Tracking Reveal About Exoskeleton Adaptation?

This study used Theia3D markerless motion capture to continuously track how individual participants adapted to a powered ankle exoskeleton, revealing gradual, highly individualized adaptation patterns that a group-averaged analysis alone would have missed. If your work involves studying human-exoskeleton interaction or occupational assistive device adoption and you want to track adaptation at the individual level, talk to our team about using Theia3D for your own longitudinal movement research.

Recent Posts