Sit-to-stand (STS) transfers are a fundamental activity of daily living, but become markedly challenging after total knee arthroplasty (TKA). Although TKA effectively relieves pain, many patients exhibit persistent quadriceps weakness and altered biomechanics, which unloads the operated limb and over-reliance on the contralateral side (Mizner and Snyder-Mackler, 2005). Similar mobility limitations and widespread use of lower-limb orthoses are reported in neurological and age-related conditions such as hemiplegia, cerebral palsy, diplegia, and frailty (Dereshgi et al., 2021; Balkman et al., 2022). Early restoration of symmetric STS is therefore a primary goal of rehabilitation.

Post-operative rigid braces stabilize the joint but offer no active torque (Barrera Sánchez et al., 2022). Users therefore adopt compensatory strategies such as trunk flexion or arm push-off (Kralj et al., 1990; Kotake et al., 1993). Active knee orthoses (AKOs) have the ability to augment knee extension during high-torque tasks (Kim et al., 2015; Shepherd and Rouse, 2017; Zhang et al., 2021). Typical STS moments of ≈ 0.4 - 0.9   N m   k g − 1 (Sibella et al., 2003; Kotake et al., 1993) often exceed what compact actuators can deliver continuously, so intelligent control is essential. Experimental studies confirm meaningful off-loading: Choi et al. reported a 19 % quadriceps average rectified EMG value (ARV) reduction when torque assistance was triggered at four tested time points (Choi et al., 2021); a self-aligning rigid AKO reduced peak electromyography (EMG) by 29 % in a post-stroke case (Sarkisian et al., 2022). Nonetheless, most evaluations involved healthy participants, bilateral or tethered prototypes, and rarely analyzed variability or user adaptation to the workload.

Impedance control modulates stiffness and damping online (Huo et al., 2022; Liu et al., 2017; Villa-Parra et al., 2017). When gains scale with real-time EMG, the assistance becomes effort-proportional (Karavas et al., 2015). Choi et al. (2021) reported ≈ 19 % reductions in quadriceps ARV during step-up and STS in healthy individuals using EMG-triggered knee assistance, while Sarkisian et al. (2022) demonstrated ≈ 29 % peak EMG reductions and improved comfort in a post-stroke case with a self-aligning powered knee orthosis. Most recently, Gunnell et al. (2025) showed that an EMG-controlled powered knee exoskeleton reduced peak quadriceps EMG by ≈ 32 % and increased affected-side knee torque by ≈ 59 % in stroke survivors when providing up to 0.5   N m ⋅ k g − 1 of assistive torque. Together, these findings establish proportional EMG-based impedance control as a promising strategy for knee support during STS. We adopt this paradigm to deliver effort-proportional stiffness without mode switching.

Repeated-measures training with a powered exoskeleton resulted in progressive decreases in EMG over 10–15 sessions, indicating motor adaptation (Kim et al., 2021). Short familiarity protocols reduced NASA-TLX workload to 34 and increased System Usability Scale scores by more than 30 % among first-time users compared to untrained users (Lau and Mombaur, 2022). Acceptance studies emphasize an unobtrusive design to avoid abandonment among older adults (Shore et al., 2022).

Bespoke, high-end exoskeletons or bulky do-it-yourself (DIY) solutions dominate current evidence. From a translational perspective, many powered knee orthoses remain prohibitively expensive and bulky, relying on custom frames and specialized actuation hardware. In contrast, retrofitting CE-certified postoperative braces with off-the-shelf actuators promises lighter, modular, and more affordable devices at the cost of reduced torque capacity. Whereas Gunnell et al. (2025) focused on high-torque assistance in stroke survivors using laboratory-grade hardware and high-fidelity EMG, we investigate whether a compact, retrofitted brace with lower torque capacity ( ≈ 0.23   N m ⋅   k g − 1 ) and consumer-grade EMG can still provide meaningful unloading in healthy users during STS. We thus view our work as complementary, targeting low-cost hardware and minimal calibration as prerequisites for future translational studies. Using ARV features as a proxy for muscle demand, we report a pilot study in healthy subjects to determine whether a budget-friendly solution reduces effort while largely preserving natural STS kinematics. Quantitative gait and joint-angle analyses to confirm kinematic fidelity will be addressed in future work.

The present work addresses this gap with the following contributions:

We retrofit a compact, CE-certified postoperative knee brace with low-cost off-the-shelf actuators and dry-electrode EMG bracelets, yielding a back-drivable, ≈ 1.1   k g active knee orthosis that requires only a brief ( ≈ 10   s ) per-user calibration.

Together, these findings demonstrate that a budget-conscious, minimally calibrated EMG-impedance controller can meaningfully unload knee extensors during STS while preserving natural coordination, thereby motivating future clinical studies in postoperative and neurologically impaired populations.

Our back-drivable active knee orthosis (Figure 1 center) is based on a modified GENUDYN®CI STEP THRU knee orthosis [Sporlastic (2023), Nürtingen] retrofitted with an AK80-9 brushless actuator [gear ratio 9:1, CubeMars (2023), Nanchang]. An MD80 v2.1 controller [MABRobotics (2025), Poznań], capable of 18 Nm peak and 9 Nm continuous torque, provided position, velocity, and impedance control via integrated encoders. The on board controller firmware limits peak torque to 16   N m . In the unpowered state, the AK80-9 exhibits a measured backdrive torque of approximately 0.5   N m CubeMars (2023). When operated in the MD80 controller’s built-in transparency mode, our measurements confirmed a mean residual reflected friction of below 0.05 ± 0.03   N m , ensuring effectively back-drivable behavior. The total device mass, including the actuator, brace, and custom mount, was mass ≈ 1.1   kg , equally shared between the M-size orthosis ( 580   g ) and motor + brace ( 550   g ) . This is comparable to lightweight tethered knee exosuits with offboard actuators ( ≈ 1.1 − 1.7   k g on-body mass, actuators not included) (Witte et al., 2017; Park et al., 2020; Yu et al., 2022) and substantially lighter than high-torque portable knee exoskeletons ( ≈ 3.6   k g including electronics) by Gunnell et al. (2025). The bill-of-materials cost of the prototype is on the order of ≈ 1500 € excluding labour (orthosis: ≈ 800 €; motor + driver ≈ 700 €), and assembly can be completed within less than 30 min using standard workshop tools. Although this still requires specialized actuators, it is substantially lower in both cost and complexity than bespoke multi-DoF exoskeletons reported in the literature, which typically rely on custom frames and actuators (Shepherd and Rouse, 2017; Zhang et al., 2021; Sarkisian et al., 2020). Typical peak knee extension moments during STS range approximately from 0.4 – 1.5   N m ⋅   k g − 1 , depending on task setup and population (Kotake et al., 1993; Sibella et al., 2003). With a 16   N m torque ceiling and our cohort’s mean body mass of 70.8   k g , our device’s nominal assistance ( τ norm = 0.23 ± 0.04   N m ⋅   k g − 1 ) , spans roughly 15 − 57 % of reported STS moments, depending on the reference value within this broad range.

For heavier users or faster, more explosive transfers, this torque ceiling will be reached earlier, limiting the achievable unloading. The custom 3D-printed brace (see Figure 1 right) connects the actuator via a linear push-rod, accommodating the orthosis’s polycentric knee joint and preserving natural kinematics. The source files can be downloaded from Scheidl et al. (2024). The system was operated using a Raspberry Pi 4B with headless Debian-based software (Raspbian OS, Raspberry Pi Ltd, 2025), communicating wirelessly via UDP with our external signal-processing pipeline.

Ten able-bodied adults (age = 34.5 ± 13.8 y (4:female; 6:male), height = 171.9 ± 8.2 cm, weight = 70.8 ± 10.4   k g , BMI = 23.8 ± 2.2   k g .   m − 2 ) were recruited. Exclusion criteria for this study included: untreated injury, neurological disorders, and cardiovascular contraindications. All volunteers provided written informed consent, and the study was approved by the Friedrich-Alexander Universität ethics board (Ref.23–350-B).

Participants executed a Sit-to-Stand ⇌ Stand-to-Sit paradigm and returned to the starting position as displayed in Figure 3 under three successive conditions: No-Ortho, Ortho-OFF, and Ortho-ON.

No-Ortho: baseline condition without the orthosis, representing the participant’s undisturbed knee biomechanics.

For every condition, each subject performed 15 uninterrupted STS cycles with their arms crossed in front of their chest on a rigid, armless chair (seat height: 47 cm). The order of conditions was fixed (No-Ortho → Ortho-OFF → Ortho-ON) to gradually introduce the device; consequently, potential learning or fatigue across blocks cannot be ruled out and is reported as a limitation. A digital metronome set to 20 beats per minute (20 bpm: 3 s inter-beat interval) provided auditory cues. At each beat, the participant was instructed to initiate either the concentric (UP) or eccentric (DOWN) transfer, depending on the preceding stable resting state (SIT or STAND). The four resulting phases are thus defined and cycled through as SIT → UP → STAND → DOWN (Schenkman et al., 1990).

The two baseline measurements (No-Ortho, Ortho-OFF) allow quantification of natural performance and the passive mechanical burden introduced by the device, respectively; these serve as references against which the EMG-assisted Ortho-ON condition is evaluated.

Table 1 summarizes phase-specific changes in median bilateral log-ARV. Side-specific responses (left and right legs) and variance analyses are detailed in the Supplementary Material, Supplementary Tables S1–S5. Medians were reported to minimize sensitivity to outliers, with dispersion visualized through boxplots (Figures 4, 5).

Participants predominantly had limited prior exposure to orthoses and EMG systems (Supplementary Material: Supplementary Table S6). Exercise frequency was high, primarily endurance-based, reflecting a physically active cohort with minimal specialized robotics experience.

NASA-TLX results (Figure 9) indicated modest overall workload without significant differences between conditions (Wilcoxon signed-rank tests, all p > 0.05 ). Performance ratings were consistently highest but showed a downward trend with orthosis use, indicating minimal perceived disruption by the device.

Participants rated orthosis usability positively (75–79 mean scores), with intuitive control perceptions remaining stable. Despite minor decreases in intuitiveness under active conditions, overall task complexity remained low. Visual and acoustic feedback was highly valued (82–87 mean scores), with a slight preference for more device-generated feedback over supervisor cues. Discomfort remained moderate, with slight increases during active power, reflecting acceptable comfort levels.

These findings align well with the intended function of an EMG-proportional impedance controller, which modulates assistance based on measured muscle activation. The observed lower median EMG amplitudes under powered conditions suggest that the orthosis successfully replaced part of the biological muscle torque, effectively reducing the muscular effort required by users. Side-specific analyses support this interpretation. Across sit, ascent, and descent, the braced right leg consistently showed larger ARV reductions than the unbraced left leg (UP: − 14.8 % vs. − 7.2 % ; DOWN: − 20.7 % vs. − 10.0 % ; Supplementary Material: Supplementary Tables S1, S2), which implies an ≈ 8 % and ≈ 12 % reduction, respectively, in the right-to-left ARV ratio compared with No-Ortho. Within the limits of EMG as a proxy for neuromuscular effort, this pattern indicates preferential unloading of the assisted limb without contralateral overuse. Because no ground-reaction forces or centre-of-pressure data were collected, these EMG-based symmetry measures reflect relative effort redistribution rather than precise mechanical load sharing between limbs. Although the reduced variance in the right leg during descent points toward more consistent muscle activation patterns, this finding alone does not definitively indicate improved stability or movement quality. Confirming stability improvements would require additional kinematic or kinetic evidence, such as detailed joint angles, segment velocities, or ground reaction forces. Consequently, EMG variance should be viewed as descriptive and indicative of neuromuscular behavior, rather than as a direct measure of functional stability or control quality.

The side-specific results further argue against contralateral compensation. The braced right leg exhibited reduced activation during sit, ascent, and descent under powered conditions. Notably, the left leg did not show a compensatory increase. Instead, it demonstrated reduced activation during ascent (powered vs. unpowered) and descent (powered vs. baseline). This consistent bilateral pattern indicates that assistance did not shift demand to the contralateral limb. Additionally, the left-leg reduction during descent suggests bilateral unloading effects, likely due to interlimb coordination rather than direct mechanical support.

Phase-specific contrasts shed light on how device mechanics interact with varying task demands. During the stand phase, the unpowered orthosis significantly increased bilateral EMG compared to baseline, an effect that active powering of the orthosis successfully mitigated. This suggests that the unpowered condition introduced additional physical demands, likely due to factors such as device weight, residual friction, or alignment constraints, which the powered assistance partially offset. In contrast, during the sit phase, the powered orthosis significantly reduced bilateral EMG, highlighting that its supportive effects extended beyond dynamic movements alone. Furthermore, during descent, both powered and unpowered conditions resulted in decreased bilateral EMG compared to baseline, with the powered condition providing a notably greater reduction. These observations underline the orthosis’s role in effectively supporting muscular effort across various phases and tasks.

Exploratory analyses offered further insights into user adaptation patterns and design implications. We observed a significant increase in within-session torque during the sit phase ( Δ = 0.084 , t ( 9 ) = 2.98 , p < 0.05 ) , although median bilateral EMG drifts remained small and statistically insignificant during both sit ( + 3.6 % , p adj = 0.134 ) and ascent ( − 5.7 % , p adj = 0.635 ) . These findings suggest modest adaptation by users across repeated trials, though notable EMG-level changes within sessions were not clearly apparent. Orthosis power utilization was substantially, 73.7 ± 19.4 % during descent and 58.5 ± 8.6 % during ascent, indicating effective engagement of the actuator. Given the device’s torque ceiling of 16   N m , it is certain that saturation occurred during more demanding descent phases, potentially limiting further muscular unloading. Consequently, enhancing the device’s torque density emerges as an important consideration for future orthosis design. While the protocol comprised only 15 repetitions per condition, the early-late comparisons and trial-wise trends (Table 2; Figures 6, 7) provide some insight into short-term adaptation. Participants increased anticipatory torque during SIT but showed only small, non-significant drifts in median ARV across repetitions. This pattern is more consistent with strategic adjustment to the EMG-driven assistance than with pronounced fatigue. However, the short exposure and healthy cohort preclude strong conclusions about long-term robustness or training effects. Multi-session protocols, ideally in clinical populations, will be needed to quantify how muscle fatigue and motor adaptation evolve over days or weeks of use.

Subjective workload, assessed using the NASA-TLX, showed no significant differences between conditions across all subscales (Figure 9). Given the small sample size and the short, highly structured protocol, the study was underpowered to detect potentially subtle subjective benefits such as reduced perceived effort during the dynamic phases. At the same time, the absence of increased workload in the Ortho-ON condition is encouraging, suggesting that the EMG-driven impedance controller can reduce extensor EMG without introducing a noticeable cognitive or physical burden. We expect that longer, more ecological usage scenarios (e.g., repeated STS as part of daily activities or rehabilitation sessions) may reveal clearer subjective benefits, particularly for users with pronounced extensor weakness.

To place our findings in the context of recent EMG-controlled knee exoskeleton work, we next compare our results with those of Gunnell et al. (2025). This research reported significant reductions in peak quadriceps EMG (32%) and substantial increases in knee extension torque (59%) in stroke survivors using an EMG-controlled powered knee exoskeleton with a torque limit of 0.5   N m ⋅   k g − 1 . In our healthy cohort, we observed more modest reductions in median ARV (11%–21% depending on phase and leg), which was not always fully exploited during ascent (mean utilization ≈ 59 % ; Figure 7), with a device whose nominal torque capacity ≈ 0.23   N m ⋅   k g − 1 ) is roughly half of that used by Gunnell et al. (2025). This is in line with their research, which achieved a ≈ 32 % reduction in peak quadriceps EMG. Together, these findings suggest a roughly dose-dependent relationship between available assistive torque and achievable EMG reduction, modulated by user adaptation and task demands. Moreover, Gunnell et al. (2025) employed high-frequency EMG sampled at 2000 Hz and targeted a single paretic muscle, whereas our controller relies on consumer-grade dry electrodes at 200 Hz and a global thigh activation pattern. This difference likely affects the precision of activation timing and amplitude estimates. Nevertheless, both studies consistently show that proportional EMG control can reduce extensor effort without inducing detrimental compensations. Our results extend this evidence to a low-cost, minimally calibrated brace, indicating that meaningful unloading may be achievable even with reduced torque capacity and simpler sensing hardware, rather than competing with high-torque custom-built braces.

The fixed order of testing conditions introduces potential sequence or learning effects, possibly influencing within-session comparisons. To mitigate this risk analytically, we explicitly modeled trial progression in the linear mixed-effects models and contrasted early versus late repetitions (Table 2). While modest adaptations were observed, such as increased preparatory ARV and torque in the SIT phase, these changes did not differentially affect any orthosis condition, and Condition × Epoch interactions remained non-significant. Nevertheless, future studies should adopt counterbalanced or randomized designs to decouple device effects from sequence-related adaptation or fatigue fully.

Although consumer-grade dry-electrode hardware simplifies deployment, it is more susceptible to motion artefacts, electrode shift, and sweat-induced impedance changes than laboratory-grade adhesive EMG systems. Yet our choice of sensors aligns with the orthosis’s objective of low cost and practicality. Additionally, our ridge-regression controller and the coarse, single-DoF task reduced the impact of such noise, but long-term robustness under daily-life conditions remains to be demonstrated. Again, here we anticipate that the quick, easy calibration allows immediate updates to the regression model weights when needed.

A further limitation concerns the specificity of the EMG-driven control signal. Because the Myo band spans the anterolateral to anteromedial thigh, the recorded envelopes primarily reflect global quadriceps activity, but inevitably also contain cross-talk from biarticular muscles such as rectus femoris and from adjacent hip musculature. Consequently, y ̂ is best interpreted as a single-leg, sagittal-plane activation measure rather than a pure knee-extensor channel. For the paced STS task studied, where both legs act in phase and extensor demand on the braced limb dominates, a global signal is appropriate and has also been used successfully other studies (Lyu et al., 2019). In more dynamic, cyclic tasks such as walking, quadriceps and hamstring muscles contribute in a phase-dependent manner to both hip and knee motion (Mohammadyari Gharehbolagh et al., 2023; Akl et al., 2021). In those settings, a purely monotonic mapping from a single global thigh signal to knee stiffness would likely be suboptimal. EMG-based gait exoskeletons therefore typically combine muscle-specific EMG features with gait-phase estimation (Chen et al., 2023; de Miguel-Fernández et al., 2023).

Similarly, the orthosis torque ceiling of 16 Nm ( ≈ 0.23   N m ⋅   k g − 1 ) may restrict effectiveness for heavier users or faster, more dynamic transitions. The small sample size and healthy cohort constrain statistical precision and generalizability. Given the relatively slow and predictable dynamics of paced STS transfers, the ≈ 250 − 300 ms rise time from EMG onset to reaching the 16 Nm torque ceiling did not manifest as perceptible delay or instability, as supported by the absence of abnormal kinematic patterns and the lack of increased subjective workload in the Ortho-ON condition. This is further illustrated by the representative time series for a powered trial (Supplementary Figure S5), where the EMG envelope of the braced thigh rises before the prediction y ̂ , yet elicited torque and EMG stay largely synced. The temporal smoothing also avoids abrupt torque jumps while still allowing participants to elicit substantial support during an extensor burst UP motion. Together with the high utilisation levels observed during the DOWN phase, this suggests that users quickly learned to exploit the compliant, EMG-scaled stiffness behaviour.

Our findings provide clear translational implications for future orthosis development. The present prototype still requires integration and is not a plug-and-play consumer product. Our use of an off-the-shelf brace and actuators primarily reduces material cost and facilitates replication by other laboratories, rather than eliminating professional involvement. We therefore frame the device as a low-cost research platform and a potential template for future industrial designs, rather than as an immediately deployable clinical solution. Effective everyday support for sit-to-stand transitions demands higher peak torque capacities without significantly increasing device weight, particularly during the more challenging descent phases. Enhanced mechanical alignment, ergonomics, and design improvements are critical to minimizing the static costs identified in unpowered conditions. Incorporating joint tracking systems, such as inertial measurement units (IMUs), would facilitate continuous joint kinematic monitoring and adaptive state estimation, essential for refining impedance control and enhancing stability assessment. Future studies should adopt randomized or counterbalanced experimental designs and integrate multiple synchronized sensor modalities, including EMG, IMU, and kinetic measurements. Incorporating metabolic and functional outcome assessments will determine whether the EMG reductions observed translate into tangible benefits, such as reduced joint loading, maintained coordination, and lower energy expenditure. Comparative studies against passive braces and other control strategies are necessary to clarify the distinct mechanisms and advantages of active assistance. Additionally, clinical trials involving diverse populations, especially those with motor impairments, are essential to comprehensively evaluate the orthosis’s clinical relevance, usability, and safety. Moreover, the present study evaluated the EMG-driven impedance controller only during paced sit-to-stand transfers in healthy adults. Systematic testing of its versatility across walking, stair ambulation, and everyday transfer tasks, as well as in clinical populations, will be crucial to establish its broader clinical applicability.