Mirkov, M., Štrbac, M., and Vujaklija, I., “Towards Minimal Sensor Setups and Calibration Protocols for Smart Mechatronic Ankle-Foot Orthoses Control,” accepted peer-reviewed extended abstract for publication in the proceedings of the International Conference on NeuroRehabilitation ICNR 2026.

Abstract: In rehabilitation robotics, protocols often require patients to perform multiple functional activities with complex setups to configure devices, which can increase fatigue and prolong setup time. We examined how setup simplification through reduced sensing modalities and fewer sensors affects performance. Using high-density EMG (HD-EMG) and inertial measurement units (IMUs) data from five subjects performing eight activities, we evaluated multiple sensor configurations for activity recognition using subject-specific, repetition-based cross-validation. Results reveal a modality–task trade-off: IMUs are well suited for ambulation-focused exercises, while other activities also require the fine information provided by EMG. This supports simpler assessment protocols with lower setup burden while retaining clinical utility in smart mechatronic ankle-foot orthoses (SMAFO) applications.

I. Vujaklija, J. Do, M. Moscu, and M. Mirkov, “Towards Context-Aware Multimodal Interface for Truly Wearable Mechatronic Ankle–Foot Orthosis Control,” in Proceedings of the International Conference on NeuroRehabilitation (ICNR 2026), 2026.

Abstract: Wearable ankle-foot orthoses demand intuitive, low-latency, and truly wearable human–machine interfaces. This work investigates a high-fidelity, multimodal interface combining HD-EMG and IMUs for context estimation on embedded hardware. Multimodal data were collected from six participants performing treadmill walking and running, squatting, sit-to-stand, and step-up tasks, including weighted variants. A complete signal-processing and LDA-based classification pipeline was implemented on a low-power microcontroller, achieving ~82% accuracy and 156ms average inference time. Results show robust discrimination between activity families and highlight challenges in separating biomechanically similar and weighted tasks.

J. Flack, M. Wu, and T. Verstraten, “Enabling Flexible Shaft Remote Actuation in Wearable Robotics via a Novel Conduit Design,” in Proceedings of the International Conference on NeuroRehabilitation (ICNR 2026), 2026.

Abstract: Flexible shafts are a promising method of remote actuation for wearable robotics but suffer from nonlinear torque transmission due to coiling. This work presents the design and validation of a novel sensorized shaft conduit that enables implementation of flexible shafts in wearable robots. The novel planar conduit design allows planar bending while limiting coiling, allowing linear torque transmission. Additionally, the custom pin-in-slot geometry of the modular conduit links reduces parasitic forces in the conduit structure during bending. Finally, a wire-draw sensor exploits the constrained bending to measure the bend angle. Use of the planar conduit was shown to reduce torque losses to 9.9% and 15.4% and increase the linearity of the transmission ( R2 = 0.984 and R2 = 0.969 ) for positive and negative torsion, respectively. The sensor was validated during human walking with a strong agreement with motion capture data (RM SE = 1.03◦, R2 = 0.997).

Abstract: Research and development in the field of lower limb orthoses focus on improving mobility in individuals with gait impairment. However, device abandonment remains high, suggesting that there is a disconnect between functionality and user adoption. In this study, we conducted a literature review, interviews, and a registry, to explore factors and gather insights into the design, prescription, and use of orthoses across stakeholders. Results show that while manufacturers focus on isolated factors and offer a limited set of design options to cover a wide range of walking impairments, disability interferes with multiple aspects of a patient’s life, and these are interconnected. Therefore, to support long-term use and satisfaction, manufacturers should account for a broader range of aspects (epidemiologic, biomechanical, psychosocial, activities, and level of functional independence) to better align the device functionality with users’ functional deficits and needs.

A. Silanos, E. Saman, V. Hörig, J. González-Vargas, and G. Severini, “Ankle–Foot Orthoses: State of the Art and Market Analysis,” in Proceedings of the International Conference on NeuroRehabilitation (ICNR 2026), 2026.

Abstract: Neurological disorders such as stroke or cerebral palsy, severely af-fect gait and frequently lead to impairments in the ankle – foot complex, re-ducing quality of living. To address these mobility issues, Ankle-Foot Orthoses (AFOs) are usually prescribed, as they maintain proper joint alignment, pro-vide stability and torque assistance. Based on their functionality, AFOs are classified into passive, active and semi-passive. Most products are passive de-vices, which are not actuated and hence lightweight and easier to use in daily life. In contrast, semi-passive and active systems include actuated mecha-nisms, which increase their costs and bulkiness. Therefore, these devices are mainly found in research, with only a few options in the market. This shows the lack of user‑centred AFOs that are adaptable, lightweight, and comfortable. This review aims to give an overview of the current AFOs state of the art and of the commercially available solutions, highlighting the main challenges in the field.

Moscu M., Kostic M., Vujaklija I., “Sequence-Aware Feature Fusion Improves Ankle-Angle Estimation Robustness in Smart Ankle-Foot Orthoses Applications,” in Proceedings of the International Conference on NeuroRehabilitation (ICNR 2026), 2026.

Abstract: Reliable intent estimation for smart ankle–foot orthoses must withstand real world perturbations. We test whether adding short horizon temporal context improves robustness in multimodal ankle angle regression from high density EMG and IMU signals across relevant rehabilitation activities. Specifically, we compare a multilayer perceptron (MLP) that fuses modality embeddings with an otherwise identical model augmented by a 1-D temporal convolution (MLP-TC). Using data from four participants, robustness is quantified as the relative change in prediction error from a clean, in distribution baseline under two conditions: (i) additive sensor noise at two levels and (ii) activity domain shifts via testing on unseen activities. Results show that temporal context consistently mitigates noise induced degradation – MLP-TC exhibits greater error stability than MLP at both noise levels – while offering inconsistent protection against cross activity shifts, where performance still declines. These findings establish a clear baseline and support incorporating temporal context to enhance noise robustness, while highlighting domain generalization across rehabilitation tasks as an open challenge.