Knee osteoarthritis (OA) is one of the most prevalent joint diseases worldwide, and mechanical loading plays an important role in its development and progression. Accurate knee joint kinematics are essential for computational models that estimate biomechanical parameters such as joint contact forces. Fluoroscopy enables accurate measurement of joint kinematics and may help overcome some limitations of conventional approaches, including soft tissue artefacts associated with optical motion capture and simplified model assumptions about knee joint motion.
This study aimed to develop and evaluate a modelling framework integrating fluoroscopy-derived knee kinematics into an existing musculoskeletal (MSK) model to assess whether this improves the prediction of tibiofemoral (TF) forces and medial-lateral load distribution, and how these outcomes are affected by different modelling choices.
Simulations were conducted using a generic MSK model and walking and squatting data from two subjects from the CAMS-knee dataset. TF flexion–extension (FE) rotations and anterior–posterior (AP) and superior–inferior (SI) translations were derived from fluoroscopy. FE rotation was prescribed during inverse kinematics (IK), while AP and SI translations were implemented as functions of FE rotation by replacing the model’s original coordinate-coupling functions. Subject-specific mean kinematic relationships were also constructed for each activity. Different kinematic prescription configurations were evaluated. Joint loading was estimated using the rapid muscle redundancy (RMR) solver. Total TF forces, as well as medial and lateral compartment forces, medial force peaks, impulse, and medial load ratio (MLR), were evaluated against in vivo measurements.
Fluoroscopy-derived kinematics showed larger excursions and distinct absolute magnitudes than the original model parametrisations. Their prescription led to changes in predicted TF forces and compartmental load distribution. For walking, the kinematic prescription had a limited influence on TF force predictions overall, although the FE-SI configuration showed the poorest agreement with in vivo data. For squatting, all configurations overestimated TF forces, but the FE-AP and FE-AP-SI configurations improved agreement with in vivo measurements. The FE-AP-SI configuration yielded the smallest total TF impulse differences relative to the in vivo reference in both subjects (1.663 BW$\cdot$s and 0.377 BW$\cdot$s), whereas FE and FE-SI produced the largest deviations. Lateral TF forces were predicted more accurately than total or medial forces. Subject-specific mean prescriptions yielded results nearly identical to trial-specific prescriptions.
The effect of prescribing fluoroscopy-derived knee kinematics in a generic MSK model depended on the activity and prescription configuration. No consistent improvements in walking were observed, whereas the FE-AP and FE-AP-SI configurations improved TF force predictions during squatting. Subject-specific mean prescriptions had a negligible effect on predicted forces, supporting a simplified implementation of the framework. Overall, this approach may improve the biomechanical fidelity of MSK models and advance understanding of knee joint loading.

Introduction
In individuals with spinal cord injury (SCI), wheelchair use is the primary means of mobility and is often associated with a sedentary lifestyle and an increased risk of secondary health complications. Although increasing physical activity can mitigate some of these effects, achieving sufficient activity levels remains challenging. Upright standing and walking remain important rehabilitation goals for individuals with SCI and may contribute to increased physical activity and improved body image.

Various exoskeletal systems have been developed to enable upright ambulation. However, many powered devices rely on heavy and bulky actuators, limiting their practicality in daily life. Passive orthotic devices, such as the (Advanced) Reciprocating Gait Orthosis ((A)RGO), allow walking with crutches but are associated with high physical effort and substantial upper-body demands. The Cloudwalker project aims to make exoskeleton-assisted ambulation accessible to a broader population of individuals with SCI by reducing the effort required during RGO-assisted walking. This study supports its development by characterising ARGO Walker–assisted gait and developing a simplified biomechanical model to enable future simulation-based evaluation of assistive strategies and design choices aimed at reducing walking effort.

Aim
The aim of this study was to characterise ARGO Walker–assisted gait in terms of joint kinematics and kinetics, and to use these insights to develop an initial simplified biomechanical model of ARGO-assisted gait in OpenSim.

Methods
A healthy participant performed ARGO Walker–assisted walking trials using crutches while motion capture and ground reaction forces from both feet and crutches were recorded. Joint angles and joint moments were computed in OpenSim using inverse kinematics and inverse dynamics. In addition, a simplified ARGO Walker model was developed in OpenSim Creator and qualitatively evaluated by prescribing experimental motion and visually inspecting human–exoskeleton alignment.

Results
ARGO Walker–assisted walking was substantially slower than healthy gait, with a mean walking speed of 0.34 m/s and a stance-dominated gait cycle (66.9% stance, 33.1% swing). Kinematic analysis revealed clear effects of the mechanical constraints imposed by the device across multiple joints, with particularly a pronounced posterior pelvic tilt and increased lumbar flexion throughout the gait cycle. Joint kinetic analysis showed markedly increased net joint moments compared with healthy reference data, particularly for hip flexion, knee extension, and lumbar extension moments.

The simplified ARGO Walker model moved largely in synchrony with the human body during walking. However, remaining misalignments between the human and exoskeleton indicate that further refinement is required before simulation-based analyses can be performed.

Conclusion
ARGO Walker–assisted gait differs substantially from healthy walking due to the mechanical constraints of the device, resulting in restricted distal joint motion and increased reliance on proximal joints and crutch support. This walking strategy is associated with elevated joint moments, particularly at the lumbar region, indicating increased mechanical demands. The developed simplified OpenSim model provides an initial framework for future simulation-based investigations of assistive strategies and design adaptations, but requires further refinement before it can be used to reliably evaluate approaches aimed at improving accessibility of ARGO-assisted walking.