舌头控制FES-机器人外骨骼恢复颈脊髓损伤患者手部功能
A tongue-controlled FES-robotic exoskeleton restores hand function in users with cervical spinal cord injury
本研究提出一种非侵入式舌头控制混合FES-机器人手部外骨骼,旨在帮助严重瘫痪的颈脊髓损伤患者恢复手部功能。系统通过无创舌头-机器人界面(nTRI)进行命令控制,结合肌腱驱动的手指屈曲和FES驱动的手指伸展,使5名四肢瘫痪用户完成5种日常抓握任务。
Introduction
Spinal cord injuries (SCI) affect over 20 million people worldwide1, and cervical injuries often result in severe impairment of hand function. Hand function is frequently reported as the most desired motor capability among individuals with paralysis, as it is crucial for quality of life and self-sufficiency2. Accordingly, integrating the human sensory-motor system with rehabilitation robotics has an exceptional potential to restore hand function.
Recent advances have enabled robot movement control through biological signals originating from muscle3,4,5 and brain activity6,7,8. In parallel, developments in hand exoskeleton technology have made grasping possible for individuals with disabilities9,10,11. Other approaches have leveraged functional electrical stimulation (FES) to facilitate hand movements and provide therapeutic benefits12,13,14. However, key challenges remain in providing reliable non-invasive control input for individuals with severe paralysis, enabling safe and dexterous hand closing and opening while minimizing system complexity, and mitigating FES-induced fatigue. Therefore, addressing these challenges is critical to restore hand function for activities of daily living (ADLs), improve quality of life, and reduce the burden on caregivers15,16,17.
For individuals with cervical SCI (tetraplegia), voluntary control inputs are typically limited to body parts above the injury level, often the head and neck, reducing the number of available control commands for interaction with assistive technologies and FES triggering. Exoskeleton control is possible with user input modalities such as eye tracking18,19, voice commands20,21, brain signals6,7,8,22, or muscle signals3,4,5,23. Additionally, control interfaces for triggering FES can utilize muscle or brain signals and a range of position, pressure, and inertial sensors13,24.
Alternatively, tongue-based control interfaces (TCIs) offer reliable command-based input modality due to the availability of robust sensor technologies25,26 and the high flexibility of the tongue27, which typically remains functionally intact even after high-level SCI. TCIs also offer advantages such as unobtrusiveness and low physical effort28. It has been shown that TCIs can be effectively used to control computers, powered wheelchairs, and robotics25,26,29,30,31,32,33,34,35. Several TCIs employ an activation unit (AU) attached to the tongue for sensor activation25,26, including the inductive tongue-computer interface, where a tongue attached AU is used to interact with inductive sensors25. However, invasive attachment methods such as a medical tongue piercing for long-term use may influence user acceptance. Built on the non-invasive sensor activation method and inductive sensor pads developed by Kirtas et al.36,37, different versions of non-invasive tongue control systems have been applied to tasks such as mode-switching in assistive robots38 and on-screen target selection39. In these approaches, users manipulate a frame-integrated AU to activate the sensors, instead of using a tongue-attached AU. However, neither invasive nor non-invasive tongue-based interfaces have yet been used to trigger FES or enable multiple grasps via a hand exoskeleton.
Providing safe and effective opening and closing of all fingers while keeping the exoskeleton compact and lightweight is a challenge for current hand exoskeletons. Hand exoskeletons that provide passive finger extension using elastic elements have been proposed40,41,42. However, executing a grasp requires overcoming the elastic resistance, potentially necessitating stronger actuators40. Exoskeletons with active extension of all fingers can provide improved dexterity and control but may become bulky due to the additional actuators. They may also pose a risk of finger hyperextension or joint misalignment if not carefully designed43,44,45,46, which can limit their safety, practicality, and effectiveness. While soft hand exoskeletons and tendon-driven mechanisms are becoming popular for their potential to reduce the size11, they rarely provide independent actuation of all fingers and many available systems offer hand closing only47,48,49.
FES offers an alternative or complementary actuation modality by activating the user’s own muscles to enable movements. In addition, FES provides therapeutic benefits including muscle strengthening, reduced muscle stiffness, increased bone density, and improved aerobic fitness50,51. However, FES can cause rapid fatigue due to unnatural recruitment order of motor units52, which may limit sustained use in functional tasks, and care must be taken to prevent FES-related skin irritation.
Hybrid upper limb systems that combine FES and robotic assistance have therefore been proposed to leverage the complementary strengths of both approaches and promote neuroplasticity and rehabilitation53,54,55,56,57,58. While reviews of such systems have highlighted their promise, they have also emphasized the need for improved tunability, reduced size, and intuitive and unobtrusive user-intent-driven control53,56. While existing hybrid approaches often employ shared actuation of the same joints in hand exoskeletons, the need for both actuators and stimulation to generate overlapping motions and challenges in achieving selective stimulation patterns for multiple grasp configurations increase overall system complexity54,57,58. Furthermore, many hybrid systems rely on conventional control input modalities56, which remain inapplicable for individuals with severe paralysis. These challenges underline the need for comprehensive hybrid solutions for restoring hand function in individuals with tetraplegia.
In this study, we present a non-invasive tongue-controlled hybrid FES-robotic hand exoskeleton for individuals with severe paralysis. The proposed system employs a unique functional segregation of tendon-driven finger flexion and FES-induced finger extension, both controlled by the tongue in an unobtrusive manner. The proposed strategy enables hand opening and five ADL grasps in individuals with hand paralysis due to SCI, including long-term injuries. For command-based control, we propose an intraoral non-invasive tongue-robot interface (nTRI) based on a multilayer printed circuit board (PCB) incorporating rounded-square inductive sensors37 and a frame-integrated activation unit (AU)39 to select multiple continuous control commands from a dedicated command panel developed for this study. This interface is applied to a tendon-driven soft hand exoskeleton providing coordinated, independent flexion of all five fingers and capable of performing multiple grasps. Finally, we introduce a tongue-controlled FES-based finger extension strategy that complements tendon-driven flexion, enabling targeted neuromuscular activation while limiting stimulation-induced fatigue and reducing overall system complexity (Fig. 1, Supplementary Movie 1).
a Overall system incorporating the nTRI for command-based control, hand exoskeleton for different grasps, and FES-induced hand opening. Further details on the system components and their implementation are provided in the Methods section. 3D rendering elements were created in Blender using publicly available assets (e.g., BlenderKit) under royalty-free licenses. b The nTRI allowed users to manipulate a frame-integrated AU across a pad consisting of 15 inductive sensor coils using tongue movements to control the system. c Hand exoskeleton incorporating five individually actuated tendon-driven fingers to enable different grasps used in ADLs and FES-based opening. d Passive opening exoskeleton relying on extension wires placed on the dorsal side of the glove fingers which was developed for comparison. In this configuration, finger extension is achieved through passive elastic elements. e Sensor data from nTRI were wirelessly transmitted to a central unit (CU), which then relayed the information to a computer via serial port. Based on activated sensors, the computer sent control commands to actuators housed in the exoskeleton control units, which pull the tendons to generate finger flexion (illustrated with turquoise dashed lines along the inner side of the glove). Thus, tendon-driven finger flexion facilitated user-selected grasps to pick up a strawberry, business card, bottle, electric toothbrush, and a pointing gesture. With the nTRI, users could also trigger FES, which was delivered by electrical stimulators, to extend their fingers (outer side of the glove is shown) and release a grasp.
Five users with tetraplegia participated in the experiments conducted over two sessions. Two configurations of the hand exoskeleton were evaluated: an FES opening exoskeleton and a passive opening exoskeleton. Functional grasping tests were performed to evaluate the systems in performing ADLs and grasp force tests were conducted to measure forces exerted by the exoskeletons. In addition, the participants performed baseline tests without wearing an exoskeleton. At the end of the second session, the participants were asked to fill in a questionnaire to evaluate the usability of the systems.
Results
Hand function in ADLs without the exoskeleton
Five users with tetraplegia (two females and three males; mean age: 40.8 ± 12.3 years; time since injury: 0.5–31.1 years, mean=13.1; Table 1) were recruited for the experiments. The inclusion criteria included the inability to grasp a bottle with the right hand.
During the baseline tasks in the first session, participants attempted to perform grasping tasks used in ADLs with different objects using their right hand without wearing an exoskeleton. Baseline performance was analyzed qualitatively based on video observations. Overall, baseline assessment showed that participants had limited or no ability to perform the grasps independently. While some users were able to complete the tasks using compensatory strategies involving wrist extension (commonly referred to as tenodesis grasping), performance was inconsistent and task-dependent, and none of the participants could reliably perform all grasp types required for the experimental tasks.
Participant 1 (P1) did not exhibit observable movement in the shoulder, arm, or fingers. Therefore, an experimenter positioned the objects in a grasping position during baseline trials. Nevertheless, P1 could not grasp any of the objects (Supplementary Movie 2).
P2 demonstrated residual movement in the arm and shoulder but no functional finger movement. As a result, a conventional grasp was not possible. However, P2 managed to lift the strawberry and business card through forearm supination and wrist extension against gravity. Despite the level of the participant’s injury and ASIA Impairment Scale (AIS) grade (Table 1), this behavior is consistent with the International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI) motor assessment, which indicated preserved wrist extension on the right side (C6: M4, Supplementary Table 1). Due to the limited ability to pre-shape the hand, P2 approached the bottle and electric toothbrush from their narrower top sections. While the attempts to lift the bottle were unsuccessful due to its weight, the participant was able to lift the toothbrush through wrist extension (Supplementary Movie 3).
P3 showed movement in the arm and shoulder, but no functional active finger movement. He was able to grasp and lift the strawberry and business card by extending the wrist, consistent with residual wrist extensor activity, although limited on the tested side (C6: M1, Supplementary Table 2). P3 could not pre-shape his hand for grasping the bottle and electric toothbrush; therefore, approached the objects from above. Bottle grasping was occasionally possible by supporting its weight with the palm despite observed difficulties; however, lifting the bottle upwards resulted in slippage. In contrast, P3 was able to lift the toothbrush upwards (Supplementary Movie 4).
Similarly, P4 was able to move his arm and shoulder, but not his fingers. He was instructed not to use compensatory wrist extension strategies (i.e., tenodesis grasping) and he could not grasp any objects (Supplementary Movie 5). During baseline trials, P5 assisted the movement of the right hand using the contralateral hand. He succeeded only in grasping the business card while all other attempts to grasp the objects were unsuccessful (Supplementary Movie 6). None of the participants were able to perform the pointing gesture, as extension of the index finger was not observed in any participant.
Hand function in ADLs while using the exoskeleton
Participants performed functional grasping tasks in ADLs (Fig. 2) using the FES opening and passive opening exoskeletons. Each grasp was repeated five times per participant in both configurations, with additional attempts in case of unsuccessful trials (see Methods). Using the proposed system, all five participants were able to complete the tasks with high success. Overall success rate for all trials performed by the participants wearing the FES opening exoskeleton was 96% while it was 89% for the passive opening exoskeleton (Fig. 3a). The success rate was calculated as the proportion of successful grasps relative to total attempts.
Participants performed the grasping tasks by manipulating the AU on the nTRI a and activating buttons on a graphical user interface (GUI) b The system facilitated five grasp types: pulp pinch of a plastic strawberry, lateral pinch of a business card, transverse volar grasp of a bottle (see Supplementary Movie 1), diagonal volar grasp of an electric toothbrush, and a pointing gesture. Participants followed GUI instructions to grasp the target object, lift and hold it, place it on the desk, and open their hand. Each grasp was repeated five times per participant with both the FES opening and passive opening exoskeletons. In case of unsuccessful attempts (e.g., object drop), trials were repeated until a successful grasp was achieved (i.e., the object was securely held for 5 seconds) (see Methods for details). The participants also triggered FES by activating a button to open their hand for the FES opening exoskeleton c A user with tetraplegia grasping a strawberry with the FES opening exoskeleton d For each participant, tendon force setpoints were individualized by adjusting default values before the trials. Performance metrics were success rates of the grasps (the number of successful grasps relative to the total number of attempts) and completion times for successful trials (elapsed time between cue presentation and grasp detection), all from the second session.
a Success rates (%) of grasping trials (defined as the proportion of trials, in which the object was securely held for 5 seconds, relative to the total number of attempts). b Completion times (s) of successful grasping trials (the elapsed time between the target grasp appearance and grasp detection). Each data point of completion time represents a single successful trial, while success rates were reported as a single value per participant across all trials, resulting in one data point per participant.
The FES opening exoskeleton achieved a high success rate in the pulp pinch grasp using a strawberry (96%) with one repeated attempt by P2 (five successes out of six attempts). For the passive opening exoskeleton, lower success rates (76%) were observed, with multiple repeated attempts required for some participants (P1 required four additional attempts, while P3 and P5 required two additional attempts each). For the lateral pinch, only one additional attempt was required with the FES opening exoskeleton (96% success), whereas all trials were successful with the passive opening exoskeleton (100%). In the bottle grasp task, the FES opening exoskeleton achieved an 89% success rate, with three additional attempts in total (two by P4 and one by P5). On the other hand, the passive opening exoskeleton success rate was 83%, which was due to five consecutive failed attempts by P5 due to slipping of the bottle. All trials of the diagonal volar grasp for the electric toothbrush were successful with both exoskeletons (100% success) (Fig. 3a, Table 2).
Overall, the participants achieved comparable task completion times using both exoskeletons (Fig. 3b, Table 2). Average task completion times were 12.7 ± 7.2 s for the FES opening exoskeleton and 13.0 ± 7.6 s for the passive opening exoskeleton. Across all tasks, the median completion times ranged from 6.9–12.7 s for the FES opening exoskeleton and 8.3–12.7 s for the passive opening exoskeleton. Despite occasional longer trials caused by user or device errors, most grasping tasks were completed in under 20 s with both exoskeletons.
Throughout the trials, varying levels of assistance and support were required across participants due to the differences in their residual upper limb function. P1 required continuous assistance from an experimenter due to the inability to move the arm and shoulder. Specifically, the experimenter positioned the objects in a grasp-ready position similar to the baseline trial. In addition, P1 required higher-than-default tendon forces for all the grasps with the FES opening exoskeleton and for the bottle and toothbrush grasping with the passive opening exoskeleton. P2 performed the tasks independently using default tendon forces with the exoskeletons and did not require any external assistance. Using default force settings for all grasps, P3 and P5 supported the bottle and toothbrush using the contralateral hand to position them in a graspable orientation prior to grasp initiation. Additionally, an experimenter provided support to P5 by stabilizing the wrist to compensate for limited proximal control. P4 required increased tendon forces to grasp the bottle with both exoskeletons and to perform the diagonal volar grasp and the pointing gesture with the passive exoskeleton; however, P4 did not require external assistance from the experimenter.
Finger extension performance was assessed qualitatively based on whether the fingers and thumb extended sufficiently to release the grasped object. No participant required increased FES current amplitudes throughout the trials and no finger hyperextension was observed. The FES current amplitudes required to achieve sufficient extension for object release are summarized in Table 3. P1 (0.5 years post-injury) reported no stimulation sensation, consistent with high-level complete SCI. Finger extension adequate for object release was achieved despite apparent finger joint stiffness. For P2 (5.2 years post-injury), FES produced extension sufficient to release the objects. P3 (23.5 years post-injury) achieved sufficient finger extension for object release, although extension of the ring and little fingers appeared more limited compared to the other digits. For P4 (5.2 years post-injury), a single pair of FES electrodes produced sufficient extension of all digits to release objects after grasping; therefore, a separate pair of thumb extension electrodes were not required. Although the required stimulation amplitude was higher than for the other participants, no undesired movements such as wrist extension or forearm supination were observed. For P5 (31.1 years post-injury), electrical stimulation produced sufficient thumb and metacarpophalangeal joint (MCP) extension, while extension at the index finger and interphalangeal joints appeared limited.
Questionnaire
With experimenter assistance provided for donning and doffing in both configurations (see Methods), the participants with SCI rated the donning and doffing difficulty of the passive opening exoskeleton 4.0 ± 1.7 and the FES opening exoskeleton 2.8 ± 1.1 on a 7-point scale (Fig. 4). P5 gave both exoskeletons the lowest possible rating (1 – not difficult). The other participants rated the FES opening exoskeleton 1 or 2 points lower than the passive opening exoskeleton, showing a preference for the FES opening exoskeleton regarding the ease of donning and doffing.
At the end of the second session, users with SCI completed a custom-made questionnaire consisting of nine items to evaluate the usability of the systems with FES opening and passive opening. The questionnaire used a 7-point Likert scale and included questions regarding donning and doffing difficulties of the exoskeletons, willingness to use the exoskeletons frequently, task completion difficulties with the exoskeletons, and discomfort, pain, and fatigue caused by FES. In this Likert scale, lower scores indicate more favorable ratings.
For willingness to use the systems frequently, the average ratings were 4.4 ± 1.9 for the passive opening exoskeleton and 3.0 ± 1.4 for the FES opening exoskeleton. P1 rated both exoskeletons as 1 (lowest), while P5 rated both as 5. For the other participants, the ratings of the FES opening exoskeleton were 2 or 3 points lower than the passive opening exoskeleton, indicating a greater willingness of frequent use for the FES opening exoskeleton.
In terms of task difficulty, the passive opening exoskeleton averaged 4.2 ± 1.9, compared with 2.8 ± 1.3 for the FES opening exoskeleton. P1 and P5 rated the exoskeletons equally (1 and 4, respectively), while the other participants rated the FES opening exoskeleton 2 to 3 points lower than the passive opening exoskeleton, suggesting lower perceived task difficulty with the FES opening exoskeleton.
Regarding the perceived discomfort, pain, and fatigue associated with FES, all the participants rated the lowest (best) possible score of 1, except P4 who rated fatigue as 2. These ratings suggest that the FES delivery caused no or minimal discomfort, pain, and fatigue throughout the experiments.
Grasp force measurements
Grasp forces with and without wearing the exoskeleton were measured using a hand dynamometer (Fig. 5a, b).
Participants wore either the FES opening or the passive opening exoskeleton gloves on their right hand and placed it on a hand dynamometer measuring the exerted force from a pinch grip a and grasp b, while an experimenter controlled the exoskeletons via a gamepad, incrementally increasing tendon force setpoints from 0 to 15 N. Each participant performed three pinch and three grasp trials with both exoskeletons as well as baseline pinch and grasp trials without any exoskeleton. c Pinch forces (N) achieved with no exoskeleton, FES opening exoskeleton, and passive opening exoskeleton. d Grasp forces (N) achieved with no exoskeleton, FES opening exoskeleton, and passive opening exoskeleton.
The pinch forces exerted by the participants (Fig. 5c) were 1.8 N ± 1.1 (median = 1.6 N, IQR = 0.9–2.2 N) at the baseline measurements, 4.8 N ± 1.1 (median = 4.9 N, IQR = 4.8–5.4 N) with the FES opening exoskeleton, and 4.6 N ± 1.1 (median = 4.5 N, IQR = 4.2–5.6 N) with the passive opening exoskeleton.
A one-way repeated-measures ANOVA showed a significant effect of condition on pinch force (F(2,8) = 29.61, p < 0.001, ƞp2 = 0.88). Post-hoc pairwise comparisons were Bonferroni corrected for multiple testing and revealed significant improvements compared with baseline: the FES exoskeleton improved the pinch force by a median of 3.0 N (95% confidence interval (CI) [1.3, 4.6]; p = 0.017), and the passive exoskeleton increased the pinch force by a median of 2.9 N (95% CI [1.4, 3.6]; p = 0.007). The difference between the exoskeletons with FES opening and passive opening was not statistically significant after Bonferroni correction (p = 1.000).
For the grasp forces (Fig. 5d), the participants achieved 2.3 N ± 2.4 (median = 1.4 N, IQR = 1.4–1.7 N) at baseline, 13.3 N ± 3.1 (median = 13.3 N, IQR = 12.1–15.2 N) with the FES opening exoskeleton, and 10.9 N ± 2.7 (median = 9.9 N, IQR = 9.5–10.5 N) with the passive opening exoskeleton.
A significant effect of condition on grasp force was shown through statistical analysis (F(2,8) = 105.83, p < 0.001, ƞp2 = 0.96). Bonferroni-adjusted pairwise comparisons indicated significant improvements with both exoskeletons. The FES exoskeleton increased the grasp force by a median of 11.2 N (95% CI [7.4, 13.5]; p = 0.001), and the passive exoskeleton improved the grasp force by a median of 8.7 N (95% CI [7.4, 9.2]; p < 0.001) compared with baseline. After Bonferroni correction, the difference between the grasp forces achieved with the FES opening and passive opening exoskeletons did not reach statistical significance (p = 0.146).
Discussion
This study is the first to investigate a non-invasive tongue-based FES-robotic control paradigm for restoring hand function in individuals with severe paralysis. The proposed system incorporates three elements: a tongue interface providing reliable and robust command-based control, a tongue-controlled soft exoskeleton with five independently actuated fingers, and a tongue-FES control strategy for hand opening. The nTRI is compact, requires no attachment to the tongue, and can provide multiple continuous commands while remaining invisible. The FES-based opening system mitigates fatigue by limiting stimulation to brief activation periods during finger extension, rather than continuous use throughout the grasp cycle. Together, these components effectively and safely facilitate user-intention-driven hand opening and multiple ADL grasps in users with cervical tetraplegia.
Hybrid FES-robotic upper-limb exoskeleton systems typically rely on shared actuation where both the robot and FES act on the same joints53,54. In contrast, the proposed FES opening exoskeleton employs functional segregation: the exoskeleton provides finger flexion and FES provides finger extension. Generating functional grasps with FES requires selective stimulation of the finger flexor muscles. While recent approaches using multi-pad surface electrode arrays have demonstrated effective non-invasive selective stimulation14, they typically rely on multi-site stimulation strategies and continuous activation to maintain grasp force. More generally, sustained FES is known to contribute to muscle fatigue during prolonged use. In comparison, object manipulation with exoskeleton-assisted grasps can be maintained longer without continuous muscle activation. By restricting FES to the extension phase, which occurs intermittently during task execution, the system reduces the duration and frequency of stimulation and thus lowers the risk of fatigue. In addition, the proposed hybrid robotic-FES system enables simpler and superficial stimulation. This approach also keeps the system compact and lightweight by eliminating the need for passive extension mechanisms or additional actuators, which may increase bulk or pose a risk of finger hyperextension. The exoskeleton’s weight on the hand (125 g) is lower than that of most systems proposed9,10,11.
Without the exoskeleton, the participants were unable to grasp the objects, except for two participants who occasionally achieved limited success using compensatory strategies involving wrist extension for a few lightweight objects. However, the participants achieved high success rates of grasping tasks using the tongue-controlled FES opening exoskeleton (96%), while the success rates were lower with the passive opening exoskeleton (89%). Although response time is an important factor influencing user acceptance of assistive technologies in real-world settings, the observed completion times (on average 12 – 13 s for both configurations) include multiple components beyond device actuation, such as object placement by an experimenter, user reaction time, hand approach, and activation of the control command. In addition, completion times were influenced by differences in participant function and assistance requirements. While the completion times remain higher than those typically required in daily activities, the questionnaire results indicated generally positive user acceptance of the exoskeleton with FES opening (Fig. 4). With user training and familiarization with the nTRI and exoskeleton, these times are expected to decrease.
For all participants, tongue-controlled FES was effective in facilitating hand opening without inducing hyperextension. Even users with 20–30 years of paralysis were able to release a grasp by FES. Required FES current amplitudes and electrode numbers varied across participants. No user needed an increase in stimulation amplitude throughout the trials, suggesting that the stimulation parameters remained sufficient for grasp release. This outcome was achieved by limiting the FES application to brief activation periods during finger extension, with a maximum duration of 2 s.
Questionnaire results showed the FES opening exoskeleton was rated more favorably than the passive opening exoskeleton in wearability, willingness to use, and ease of use. The less favorable (higher) usability scores of the passive opening exoskeleton may in part reflect discomfort and increased difficulty associated with the specific prototype design, including the use of extension wires and resulting mechanical resistance. However, these findings are not intended to generalize to all passive-extension exoskeletons, but rather reflect the characteristics of the implementation used in this study. The most favorable (lowest) scores given by P1 regarding the willingness to use exoskeletons may be attributed to the severity of her impairment and low perception of task difficulty. Although stimulation amplitude, neurological level and lesion completeness, and the resulting sensory responsiveness to FES varied among participants, no participant reported discomfort or pain associated with stimulation, including P4 who used a stimulation amplitude of 32 mA. Additionally, all the participants rated fatigue as 1 (no fatigue), except for P4 (rated 2), indicating very low subjective perception of fatigue. However, these responses reflect perceived experience rather than physiological muscle fatigue and should be interpreted accordingly. Objective assessment of fatigue (e.g., through force measurements or electromyographic analysis) was not performed and represents a direction for future work.
The nTRI enabled robust and continuous control of multiple grasp types and FES-triggered hand opening, even in individuals with complete long-term SCI-induced tetraplegia. This functionality is supported by its spatial sensing performance, which enables accurate estimation of AU position. Additional characterization of the interface resolution is provided in Supplementary Information (See Supplementary Notes 1, Supplementary Fig. 1, and Supplementary Fig. 2). The current implementation supports five distinct grasps; however, the high-resolution sensor performance of the nTRI allows additional grasp types to be enabled through software-level mapping. The nTRI’s smaller size and lack of a tongue-attached AU may further enhance the usability. Although oral usability aspects of the nTRI have not yet been evaluated, a previous study with the invasive iTongue® for computer control demonstrated high usability, allowing users with tetraplegia to speak, drink, and eat while wearing the system59.
Grasp force measurements showed both exoskeletons significantly improved grasping forces in all participants compared to baseline. However, because the measured forces were generated with exoskeleton assistance, they should be interpreted as a measure of functional assistive capability rather than intrinsic grasp strength. On average, pinch forces were 2.7 and 2.6 times higher, and grasp forces were 5.9 and 4.8 times higher, relative to the no-exoskeleton condition. Although the difference between the FES opening and passive opening exoskeletons did not reach statistical significance, participants showed a trend toward higher grasp forces with the FES opening exoskeleton (mean increase of 22%). While this may be related to the absence of bending resistance from extension wires, this observation should be interpreted with caution and requires further investigation. Despite inter-subject variability, achieved pinch forces of 2.6–7.0 N would be sufficient for ADLs, exceeding reported requirements for lightweight objects60. Measured grasping forces ranging from 8.6 to 17.1 N are comparable to those provided by other hand exoskeleton systems9,10,11 and sufficient for ADLs, given reported minimum requirements of 10–13 N10. Since the exoskeleton can provide up to 20 N tendon force per finger but was limited to 15 N during testing for safety reasons, higher pinch and grasp forces could potentially be achieved.
Despite promising results, the study has limitations. The small sample size limits generalizability of the findings and the system’s portability is currently constrained by external hardware such as stimulators, which may be improved in future versions. Visual feedback was used in the experiments; however, it could be eliminated through user training61. Since the FSRs were located only at the fingertips, grasp completion often relied on measured tendon forces in addition to FSR-based contact detection. Future versions may incorporate additional sensors distributed across the hand to improve grasp detection. The system does not fully eliminate the user dependency since some assistance is still needed for donning and doffing the exoskeleton and the nTRI. A FES setup for hand opening requires careful placement of electrodes and calibration for each individual, making the initial setup potentially time consuming, but for long time use, individual electrode placements may be marked to facilitate shorter setup times. Further, potential therapeutic effects of FES rehabilitation with the system may be explored in the future.
The results show that the proposed hybrid system could restore the hand function of individuals with tetraplegia by enabling different grasps used in ADLs. Thereby, the system has the potential to improve the quality of life and reduce the burden on caregivers and healthcare systems. As the presented system is based on an adapted version of a commercial exoskeleton glove and a newly developed version of a commercial tongue-computer interface, potential industry collaborations could facilitate the translation of the study outcomes into market-ready solutions within reasonable time.
Methods
Participants
Five users with tetraplegia were recruited for the experiments through clinicians at the Spinal Cord Injury Centre of Western Denmark, with injury characteristics summarized in Table 1. For participants who demonstrated grasping facilitated by wrist extension during baseline assessment (P2 and P3), available ISNCSCI motor examination scores from clinical records were reviewed retrospectively to further characterize residual motor function (Supplementary Tables 1 and 2). Except for P4, none of the participants had prior experience with tongue interfaces or exoskeletons.
The main inclusion criteria were being between 18 and 75 years of age, reduced or absent motor function in the right arm caused by a spinal cord injury (corresponding to ASIA Impairment scale (AIS) grades A to D, where A indicates complete loss of motor and sensory function and D indicates partially preserved motor function), not being able to repeatedly use the right hand and arm to grab a bottle with a straw from a table and drink from it while in a seated position, and having at least some tongue functionality. The main exclusion criteria for the participants were pregnancy, drug addiction, psychiatric diagnoses, active implanted devices, or any implants on the right arm. All procedures conducted were in accordance with the approval of the North Denmark Region Committee on Health Research Ethics (reg. no.: N-20230044) and with the Declaration of Helsinki. Signed informed consents from participants, including consent to photography and video recording, were obtained prior to the experiments.
Non-invasive tongue-robot interface
The nTRI incorporated a sensor pad consisting of the 15 mm-wide, 12-layer PCB with 15 rounded-square inductive sensor coils arranged in a 3 × 5 grid to develop a compact and simple device allowing higher quality signals37 (Fig. 2a). The nTRI also comprised a PCB with mounted electronic components for signal processing, a 2-layer flexible PCB connecting sensors and electronics, a rechargeable lithium polymer battery (40 mAh, 10 mm × 12 mm x 3 mm), a charging coil placed around the PCBs for wireless power transfer, and a handmade wireframe structure. The wireframe featured an integrated titanium sphere serving as an AU, thereby eliminating the need for direct attachment of the AU to the tongue for sensor activation.
All components were assembled inside a mold and encapsulated using a two-component (liquid and powder) orthodontic resin (Dentsply Sirona Inc, US). The wireframe was made manually by bending nickel-free stainless steel alloy wires commonly used in orthodontics (MENZANIUM®, SCHEU-DENTAL GmbH, Germany). It consisted of two parallel wires, each with a diameter of 0.7 mm, and a 0.6 mm wire with loops at each end, which held a titanium AU with a diameter of 6 mm. The wireframe design allowed two-dimensional control of the AU using tongue movements (Fig. 2a). The nTRI measured 30.2 mm in length, 20.6 mm in width, and varied in thickness from 7.8 mm to 10.5 mm. The tongue interface weighed 9.2 g.
Signal detection with the nTRI was based on the variable induction principle which was initially introduced in ref. 62. A multiprocessor (Atmega88PA, Microchip Technology Inc., US) embedded in the PCB with electronics generated a square wave signal which was fed through a band-pass filter, formed by a 10 nF capacitor in parallel with each of the 15 inductive coils. When the AU was placed close to a sensor coil, the output voltage of the band-pass filter changed, indicating an activation until removal of the AU. The output signal was processed by a detector consisting of an amplifier, a rectifier, and a low-pass filter. Once all inductive coils were multiplexed, the data were transmitted to a central unit (CU) through 2.4 GHz radio communication via a radio chip (nRF24L01 + , Nordic Semiconductor, Norway). Afterwards, the CU relayed the sensor data to a computer via a serial port at 30 Hz. The position of the AU on the sensor pad was then estimated using the transmitted data from each sensor and the weighted average of neighbors (WAN) method63. Thereby, the users could select and activate multiple buttons designed for a GUI to control the hand exoskeleton.
Individual dental retainers, in which the nTRI was embedded, that fit the palate of each participant (as illustrated in Fig. 1b) were created before the experiments. An intraoral dental scanner was used to scan the participants’ palate and upper teeth, and the models were printed using a dental resin printer (Form 2, Formlabs Inc., US). Temporary dental retainers were made by thermoforming a 0.5 mm thick dental sheet (Duran, SCHEU-DENTAL GmbH, Germany) over the participants’ 3D printed models. Finally, the nTRI was glued onto the thermoformed dental sheets using a UV curing dental resin (CLEAR-BLOKKER, SCHEU-DENTAL GmbH, Germany) for each participant. Prior to use, the created mouthpieces were disinfected in chlorine solution.
Hand exoskeleton
The advanced five-finger soft exoskeleton glove (Fig. 1c) was an adapted version of the commercial fabric exoskeleton glove Carbonhand® (Bioservo Technologies AB, Sweden), which originally incorporated three fingers (thumb, middle, and ring). To obtain a five-finger glove that allowed greater dexterity in ADLs, two fingers from one medium sized glove for the right hand were integrated into an identical three-finger glove. Further, a dedicated Robot Operating System (ROS) based exoskeleton control system facilitating independent tendon-driven flexion of five fingers actuated by two Carbonhand® control units was developed to enable different grasp types used in ADLs. The exoskeleton glove also incorporated force sensitive resistors (FSRs) on the tip of each finger, which could be used to detect whether the fingers were in contact with an object. However, unlike tendon-driven systems with active opening mechanisms, this glove did not have dorsal-side tendon routing for finger extension to minimize size, weight, and to prevent potential safety issues such as finger hyperextension. The exoskeleton weight was 125 g on the hand, and in total the system weighed 1155 g including the control units. Based on this exoskeleton platform, two configurations differing in finger extension mechanisms were evaluated: an FES opening exoskeleton and a passive opening exoskeleton.
In the FES opening exoskeleton configuration, the exoskeleton provided tendon-driven flexion, while finger extension was achieved via FES, activating the participant’s extensor muscles to release grasps. This functional segregation allowed finger flexion and extension to be controlled independently, reducing mechanical complexity while mitigating fatigue.
For comparison, a passive opening exoskeleton was developed by integrating passive extension wires on the dorsal side of the fingers of the glove (Fig. 1d), similar to40. When the flexion tension on the exoskeleton glove’s tendons was released, the fingers could extend to their original position to release a grasp. Super-elastic Nickel-Titanium (Nitinol) wires were used for passive extension since this alloy could withstand high bending forces without plastic deformation at temperatures above the transformation temperature due to its mechanical properties64. 3D printed parts were sewn to the five-finger exoskeleton glove to guide two 0.8 mm diameter Nitinol wires on the right and left dorsal aspect of each finger. The passive extension system allowed easy insertion and removal of wires using bolts and nuts. The passive opening exoskeleton weighed 134 g on the hand.
Functional electrical stimulation
To release an object and prepare the hand for grasping, a FES system enabling finger extension was developed. Electrical stimulation was delivered using two DS5 isolated bipolar constant current stimulators (Digitimer Ltd, UK) providing a current output proportional to the shape of voltage input. Asymmetrical balanced biphasic waveforms with a pulse width of 150 µs and frequency of 30 Hz were used to stimulate finger extensors. The parameters were selected based on an exploratory pilot evaluation conducted prior to the study, as well as consistency with commonly reported values for upper-limb FES in the literature24,56,65. A NI USB-6212 DAQ device (National Instruments, TX, USA) was used to generate the voltage input. The FES application was limited only to the finger extension, and the duration of the electrical stimulation was restricted to 2 s to prevent fatigue. By limiting the maximum stimulation intensity delivered to the muscles for each participant, the safety of the system was ensured, and possible issues such as discomfort and pain were avoided.
Experimental design
The experiments were conducted on two consecutive days. Each of the two one-day sessions lasted a maximum of 3 h, with the first session dedicated to training and familiarization with the tasks and the system. The experiments were carried out at the Spinal Cord Injury Centre of Western Denmark, Viborg, Denmark. During the experiments, the users were seated in their own wheelchair in front of a table with a screen and objects (Fig. 2d).
Before starting each experimental session, FES electrodes were placed on the participants’ forearms after skin preparation using alcohol wipes. Initial electrode placement was guided by anatomical landmarks. To trigger four-digit extension, a 3.3 × 5.3 cm stimulation electrode serving as cathode (PALS® Electrodes, Axelgaard, CA, USA) was placed over the muscle belly of the Extensor Digitorum Communis (EDC), approximately along the line between the lateral epicondyle and the radial styloid. A 5 × 5 cm electrode serving as anode (Dura-Stick Plus, DJO Global LLC., CA, USA) was positioned distally near the wrist. For thumb extension, electrode placement targeted the Extensor Pollicis Longus (EPL), with the 3.3 × 5.3 cm electrode (cathode) positioned between the four-digit electrode pair and the 5 × 5 cm electrode (anode) was positioned toward the radial side of the wrist (Fig. 2c).
Initial stimulations were applied to familiarize the participants with FES and to determine electrode configurations and stimulation amplitudes that were acceptable within discomfort and pain thresholds while being sufficient to produce finger extension for grasp release. Starting from 1 mA, stimulation amplitude for four-digit extension was increased in steps of 1 mA, and the resulting movements were visually assessed. Electrode positions were then iteratively adjusted based on observed motor responses to optimize selectivity and extension quality (e.g., to minimize undesired wrist movements or improve extension of specific digits). For example, if extension of the little finger was more pronounced than that of the index finger, the cathode was shifted toward the radial side to improve activation of index-extending muscle fibers. Although this procedure was primarily aimed at producing four-digit extension, sufficient thumb extension was also achieved in some participants. In this case, the electrode pairs positioned for thumb extension were removed. For other cases, a similar electrode adjustment procedure was applied for thumb extension, including the incremental increase of stimulation amplitude. Electrode configurations and corresponding outcomes were noted during the adjustment process to allow efficient return to previously effective placements if required. The optimal stimulation amplitudes required for thumb and four-digit extension were recorded, and electrode locations were marked to reduce setup and calibration times in the second session.
After the FES setup at the first session, the participants were asked to attempt to perform all five experimental grasps (pulp pinch, lateral pinch, transverse volar grasp, diagonal volar grasp, and pointing gesture) three times without wearing any hand exoskeleton to obtain a baseline performance assessment of the experimental ADLs. In addition, they attempted to pinch and grasp a Go Direct® Hand Dynamometer (Vernier, OR, USA) without any hand exoskeleton to obtain baseline grasp force measurements (Fig. 5, a and b). These measurements were intended to evaluate functional force generation during exoskeleton-assisted grasping relative to the participants’ baseline unassisted condition, rather than recovery of voluntary motor function or intrinsic muscular strength. An experimenter donned either the FES opening or passive opening exoskeleton on the participants’ right hand, as the participants did not have sufficient arm and hand function to perform donning or doffing independently. The participants placed their hands in a grasping posture on the sensor pads of the dynamometer and were instructed not to voluntarily apply force. Next, an experimenter gradually increased the tendon pulling force setpoints from 0 to 15 N for both exoskeletons, using buttons on a gamepad (Thrustmaster Dual Analog 4). The exoskeletons could provide up to 20 N of tendon force per finger; however, the maximum applied force was uniformly limited to 15 N for both exoskeletons to assure a safe reference for testing. Force setpoints were kept at 15 N for 5 s, then set to 0 to release the tendon forces. The thumb and index finger were actuated for pinching, while all fingers were actuated for grasping. One training and three experimental trials were performed for pinch and grasp force measurements with each exoskeleton.
After the baseline and initial measurements, an experimenter placed the nTRI on the palate of the participants (Supplementary Movie 1). Before the actual grasping trials, the participants performed each grasp once to individualize the tendon force setpoints by adjusting the default values, if necessary. Then, the participants performed five trials, each consisting of the five grasps listed below, in a predetermined order. An experimenter placed the objects on the table in front of the participants. For P1, who was unable to move the arm and shoulder, the experimenter instead positioned the objects in a grasp-ready configuration. A trial was considered successful if the participant grasped, lifted, and held the object for 5 s securely without dropping. In case of failure (e.g., object slip or drop), the trial was repeated until a successful grasp was achieved.
Two users started performing tasks with the FES opening exoskeleton, while other three users started with the passive opening exoskeleton. After completing the tasks with first configuration, participants took a 15 min break while an experimenter switched the exoskeleton configuration by removing or inserting passive extension wires. Following the break, participants performed the same set of tasks with the second configuration. In both configurations, the participants followed the GUI instructions to grasp the target object, lift, and hold it for 5 s, then placed the object on the desk and opened their hands (Fig. 1a and Fig. 2d).
Assistance levels were adapted based on participants’ residual motor function. However, assistance was limited to positioning the object in a graspable configuration prior to grasp initiation, either by an experimenter or via contralateral hand use, and stabilization of the participant’s upper limb (e.g., wrist support) in cases of insufficient proximal control. No assistance was provided to support the object during the 5-second holding phase. Task completion times were measured as the elapsed times between the appearance of the target grasp on GUI and grasp detection. This duration includes object placement by an experimenter, user reaction time, hand approach, activation of the correct button, and execution of the grasp by the exoskeleton. Consequently, completion times represent overall task performance and may reflect differences in participant function and assistance requirements. Task success rates for each grasp were calculated by the ratio of the number of successful trials to the number of total trials.
The five grasp types performed in the experiment were:
-
Pulp pinch through simultaneous index finger and thumb flexion to grasp a plastic strawberry (6.5 g)
-
Lateral pinch through index finger flexion followed by thumb flexion to grasp a business card (1.4 g)
-
Transverse volar grasp through flexion of all fingers to grasp a bottle (521 g)
-
Diagonal volar grasp through flexion of all fingers except the thumb to grasp an electric toothbrush (105 g)
-
Pointing gesture through flexion of the middle, ring, and little fingers followed by thumb flexion
The participants performed the grasps by tongue-activating the relevant nTRI button reflected on the GUI (Fig. 2b) by pressing it for a dwell time of 0.8 seconds and keeping the button activated. This dwell time was selected to minimize unintended activations, while allowing responsive control. While prior work shows that 0.6 s is sufficient to avoid false activations66, a slightly longer dwell time was used to improve robustness, particularly for users without prior experience.
During grasp execution, tendon forces were increased incrementally, and commands were sent to the exoskeleton actuators through ROS to allow for gradual force application and user correction. Grasp detection was primarily achieved when the fingertip contact forces measured with FSRs exceeded a threshold of 50 mN, which was empirically determined during system development as the approximate contact force required to securely hold a bottle without slippage, and remained above this threshold for 0.33 s, at which point actuator motion was stopped. However, because the FSRs were limited to fingertip locations, the FSR-based stopping condition was not always satisfied for all grasp scenarios. In such cases, measured tendon forces were used as a secondary and complementary grasp stopping criterion. When the tendon forces reached predefined setpoints, actuators were stopped to ensure completion of the grasp without applying excessive force. Default tendon force setpoints were set at 7 N (pulp pinch, diagonal volar grasp, pointing), 10 N (lateral pinch), and 5 N (transverse volar grasp) for the FES opening exoskeleton, and at 10 N (pulp pinch, diagonal volar grasp, pointing), 15 N (lateral pinch), and 7 N (transverse volar grasp) for the passive opening exoskeleton.
Before initiating a grasp, the fingers needed to be in an extended position. Therefore, the activation of a grasp button initially triggered 2 seconds of FES to extend the fingers and thereby prepare the hand for grasp, while participants concurrently adjusted their hand position relative to the object. The pointing command did not activate FES, as no object interaction was required. The delay between the button activation and FES delivery was less than 100 milliseconds. After completing a grasp, activating the opening button by pressing the AU up with the tongue for 1 s of dwell time released all the tendon forces followed by the application of 2 s of FES to extend the fingers and thereby to release the grasp. When using the passive opening exoskeleton, activating the opening button released the tendon forces for flexion to extend the fingers as the passive extension wires returned to their original position. The dwell time for the opening command was set to 1 s, compared with 0.8 s for grasp selection commands, to reduce the likelihood of unintended activation during grasp execution, which could interrupt grasp formation before the object was securely held.
Questionnaire design
At the end of the second experimental session, participants completed a custom-designed questionnaire to assess the usability of two exoskeleton configurations (FES opening and passive opening). Participants did not have prior experience with any exoskeleton. The questionnaire was in Danish and consisted of nine items evaluating key aspects relevant to exoskeleton use, including donning and doffing difficulty, willingness for frequent use, task difficulty, and perceived discomfort, pain, and fatigue associated with FES (Fig. 4). The questionnaire was intended as an exploratory, system-specific assessment rather than a standardized or validated usability measure.
The questionnaire design was partially inspired by established usability instruments such as the Post-Study System Usability Questionnaire (PSSUQ) and System Usability Scale (SUS), while being adapted to focus on system-specific characteristics, such as donning and doffing and FES-related perception, which are not fully covered by standardized questionnaires. A 7-point Likert scale (with lower scores indicating more favorable ratings) was used for all items to provide sufficient resolution in capturing user responses while keeping the questionnaire concise and focused on relevant usability aspects.
Differences in donning and doffing difficulty were assessed, as the passive opening exoskeleton includes extension wires that maintain the fingers in an extended state, which can make finger insertion into the glove more challenging compared to the FES opening configuration. Questions related to the difficulty of donning and doffing and task execution were phrased to encourage neutral and reflective responses (i.e., “How difficult was to …”). In contrast, items assessing willingness for use were formulated as agreement statements (i.e., “I would use … frequently”), consistent with standard Likert-scale approaches for capturing user preference. The ordering of the questions enabled participants to evaluate both configurations by presenting related items consecutively (e.g., task difficulty for passive and FES opening exoskeletons), thereby facilitating direct comparison while absolute ratings allow for independent assessment of each configuration. Items related to fatigue, discomfort, and pain were included to capture user-perceived responses to FES. In particular, fatigue responses reflect perceived experience rather than objective physiological muscle fatigue.
Statistical analysis
Statistical analyses were conducted in IBM SPSS Statistics (version 29) to compare the pinch and grasp forces achieved with the FES opening exoskeleton, passive opening exoskeleton, and baseline (without exoskeleton). All data are reported as mean values ± standard deviation or median with interquartile range. The normality of distributions was assessed with the Shapiro-Wilk test and visual inspection of Q-Q plots based on the residuals of each repeated measure. The assumption of sphericity was also assessed with Mauchly’s test of sphericity.
A one-way repeated measures ANOVA was conducted to test differences among three within-subject conditions for grasp and pinch force measurements. Effect sizes for ANOVA were reported as partial eta squared (ƞp2). Following a significant main effect, post-hoc pairwise comparisons for multiple testing (FES exoskeleton vs. baseline, passive exoskeleton vs. baseline, and FES exoskeleton vs. passive exoskeleton) were conducted using paired samples t-tests with Bonferroni-adjusted p-values. The significance level was set at α = 0.05. Moreover, median paired differences with 95% confidence intervals were reported for the interpretability of observed effects.
Data availability
All data needed to evaluate the paper’s conclusion are available in the paper or the Supplementary Information.
Code availability
All code used to support the findings of this study is available from the corresponding author upon reasonable request.
References
Ding, W. et al. Spinal Cord Injury: The Global Incidence, Prevalence, and Disability From the Global Burden of Disease Study 2019. Spine 47, 1532 (2022).
Snoek, G. J., IJzerman, M. J., Hermens, H. J., Maxwell, D. & Biering-Sorensen, F. Survey of the needs of patients with spinal cord injury: impact and priority for improvement in hand function in tetraplegics. Spinal Cord 42, 526–532 (2004).
Lu, Z., Chen, X., Zhang, X., Tong, K.-Y. & Zhou, P. Real-Time Control of an Exoskeleton Hand Robot with Myoelectric Pattern Recognition. Int. J. Neural Syst. 27, 1750009 (2017).
Park, S. et al. Multimodal Sensing and Interaction for a Robotic Hand Orthosis. IEEE Robot. Autom. Lett. 4, 315–322 (2019).
De la Cruz-Sánchez, B. A., Arias-Montiel, M. & Lugo-González, E. EMG-controlled hand exoskeleton for assisted bilateral rehabilitation. Biocybern. Biomed. Eng. 42, 596–614 (2022).
Frolov, A. A. et al. Post-stroke Rehabilitation Training with a Motor-Imagery-Based Brain-Computer Interface (BCI)-Controlled Hand Exoskeleton: A Randomized Controlled Multicenter Trial. Front. Neurosci. 11, 400 (2017).
Soekadar, S. R. et al. Hybrid EEG/EOG-based brain/neural hand exoskeleton restores fully independent daily living activities after quadriplegia. Sci. Robot. 1, eaag3296 (2016).
Araujo, R. S., Silva, C. R., Netto, S. P. N., Morya, E. & Brasil, F. L. Development of a Low-Cost EEG-Controlled Hand Exoskeleton 3D Printed on Textiles. Front. Neurosci. 15, 661569 (2021).
Tran, P., Jeong, S., Herrin, K. R. & Desai, J. P. Review: Hand Exoskeleton Systems, Clinical Rehabilitation Practices, and Future Prospects. IEEE Trans. Med. Robot. Bionics 3, 606–622 (2021).
du Plessis, T., Djouani, K. & Oosthuizen, C. A Review of Active Hand Exoskeletons for Rehabilitation and Assistance. Robotics 10, 40 (2021).
Chu, C.-Y. & Patterson, R. M. Soft robotic devices for hand rehabilitation and assistance: a narrative review. J. NeuroEngineering Rehabil. 15, 9 (2018).
Duan, R. et al. Clinical Benefit of Rehabilitation Training in Spinal Cord Injury: A Systematic Review and Meta-Analysis. Spine 46, E398 (2021).
Khan, M. A. et al. A systematic review on functional electrical stimulation based rehabilitation systems for upper limb post-stroke recovery. Front. Neurol. 14, (2023).
Crema, A. et al. A Wearable Multi-Site System for NMES-Based Hand Function Restoration. IEEE Trans. Neural Syst. Rehabil. Eng. 26, 428–440 (2018).
Geyh, S. et al. Quality of life after spinal cord injury: a comparison across six countries. Spinal Cord 51, 322–326 (2013).
Lee, K. E. & Jeoung, B. An analysis of the effectiveness of rehabilitation protocols for patients with spinal cord injury: A systematic review. J. Pub. Health https://doi.org/10.1007/s10389-023-02115-9 (2023).
Backx, A. P. M., Spooren, A. I. F., Bongers-Janssen, H. M. H. & Bouwsema, H. Quality of life, burden and satisfaction with care in caregivers of patients with a spinal cord injury during and after rehabilitation. Spinal Cord 56, 890–899 (2018).
Hansen, J. P. P., Bækgaard, P., Valgeirsdottir, D. & Beier, S. Universal Design of Gaze Interactive Applications for People with Special Needs. in Proceedings of the 2023 Symposium on Eye Tracking Research and Applications 1–7 (Association for Computing Machinery, New York, NY, USA, 2023). https://doi.org/10.1145/3588015.3589666.
Meena, Y. K. et al. EMOHEX: An eye tracker based mobility and hand exoskeleton device for assisting disabled people. in 2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC) 002122–002127 https://doi.org/10.1109/SMC.2016.7844553 (2016).
Tran, P. et al. FLEXotendon Glove-III: Voice-Controlled Soft Robotic Hand Exoskeleton With Novel Fabrication Method and Admittance Grasping Control. IEEEASME Trans. Mechatron. 27, 3920–3931 (2022).
Guo, Y., Xu, W., Pradhan, S., Bravo, C. & Ben-Tzvi, P. Personalized voice activated grasping system for a robotic exoskeleton glove. Mechatronics 83, 102745 (2022).
Randazzo, L., Iturrate, I., Perdikis, S. & Millán, J.dR. mano: A Wearable Hand Exoskeleton for Activities of Daily Living and Neurorehabilitation. IEEE Robot. Autom. Lett. 3, 500–507 (2018).
Hameed, H. K., Hassan, W. Z. W., Shafie, S., Ahmad, S. A. & Jaafar, H. A Review on Surface Electromyography-Controlled Hand Robotic Devices Used for Rehabilitation and Assistance in Activities of Daily Living. JPO J. Prosthet. Orthot. 32, 3 (2020).
Marquez-Chin, C. & Popovic, M. R. Functional electrical stimulation therapy for restoration of motor function after spinal cord injury and stroke: a review. Biomed. Eng. OnLine 19, 34 (2020).
N. S. Andreasen Struijk, L. et al. Development and functional demonstration of a wireless intraoral inductive tongue computer interface for severely disabled persons. Disabil. Rehabil. Assist. Technol. 12, 631–640 (2017).
Kong, F., Sahadat, M. D. N., Ghovanloo, M. & Durgin, G. D. A Stand-Alone Intraoral Tongue-Controlled Computer Interface for People With Tetraplegia. IEEE Trans. Biomed. Circuits Syst. 13, 848–857 (2019).
Kandel, E. R., Schwartz, J. H. & Jessell, T. Principles of Neural Science, Fourth Edition. (McGraw-Hill Companies,Incorporated, 2000).
Lau, C. & O’Leary, S. Comparison of Computer Interface Devices for Persons With Severe Physical Disabilities. Am. J. Occup. Ther. 47, 1022–1030 (1993).
Ostadabbas, S. et al. Tongue-controlled robotic rehabilitation: A feasibility study in people with stroke. J. Rehabil. Res. Dev. 53, 989–1006 (2016).
Huo, X., Wang, J. & Ghovanloo, M. Wireless control of powered wheelchairs with tongue motion using tongue drive assistive technology. in 2008 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society 4199–4202 https://doi.org/10.1109/IEMBS.2008.4650135 (2008).
Johansen, D., Cipriani, C., Popović, D. B. & Struijk, L. N. S. A. Control of a Robotic Hand Using a Tongue Control System—A Prosthesis Application. IEEE Trans. Biomed. Eng. 63, 1368–1376 (2016).
Pálsdóttir, ÁA., Kæseler, R. L., Bengtson, S. H., Moeslund, T. B. & Struijk, L. N. S. A. Remote Tongue-Based Control of a Wheelchair Mounted Assistive Robotic Manipulator and the Effect of Adaptive Semi-Automation. IEEE Trans. Biomed. Eng. 72, 14–23 (2025).
Mohammadi, M. et al. Tongue control of a five-DOF upper-limb exoskeleton rehabilitates drinking and eating for individuals with severe disabilities. Int. J. Hum.-Comput. Stud 170, 102962 (2023).
Shi, L. et al. Non-Invasive Intraoral Stand-Alone Tongue Control System Based on RSIC-V Edge Computing. Appl. Sci. 13, 9490 (2023).
Hou, B., Yang, D., Ren, X., Yi, L. & Liu, X. A tactile oral pad based on carbon nanotubes for multimodal haptic interaction. Nat. Electron. 7, 777–787 (2024).
Kirtas, O., Mohammadi, M., Bentsen, B., Veltink, P. & Struijk, L. N. S. A. Design and evaluation of a noninvasive tongue-computer interface for individuals with severe disabilities. in 2021 IEEE 21st International Conference on Bioinformatics and Bioengineering (BIBE) 1–6 https://doi.org/10.1109/BIBE52308.2021.9635238 (2021).
Kirtas, O., Veltink, P., Lontis, R., Mohammadi, M. & Andreasen Struijk, L. N. S. Development of inductive sensors for a robotic interface based on noninvasive tongue control. in 2022 International Conference on Rehabilitation Robotics (ICORR) 1–6 https://doi.org/10.1109/ICORR55369.2022.9896548 (2022).
Santos Cardoso, A. S., Mohammadi, M., Kæseler, R. L., Jochumsen, M. & Andreasen Struijk, L. N. S. Assessing Mode-Switching Strategies for Assistive Robotic Manipulators Using a Preliminary Version of the Novel Non-invasive Tongue-Computer Interface. in 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) 1–4 https://doi.org/10.1109/EMBC40787.2023.10340946 (2023).
Santos Cardoso, A. S. et al. Mastering Tongue-Computer Interfaces: A Pilot Study on How Users Improve With Practice. in 2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) 1–6 https://doi.org/10.1109/EMBC58623.2025.11251588 (2025).
Tran, P., Jeong, S., Wolf, S. L. & Desai, J. P. Patient-Specific, Voice-Controlled, Robotic FLEXotendon Glove-II System for Spinal Cord Injury. IEEE Robot. Autom. Lett. 5, 898–905 (2020).
In, H., Cho, K.-J., Kim, K. & Lee, B. Jointless structure and under-actuation mechanism for compact hand exoskeleton. in 2011 IEEE International Conference on Rehabilitation Robotics 1–6 https://doi.org/10.1109/ICORR.2011.5975394 (2011).
Burns, M. K., Pei, D. & Vinjamuri, R. Myoelectric Control of a Soft Hand Exoskeleton Using Kinematic Synergies. IEEE Trans. Biomed. Circuits Syst. 13, 1351–1361 (2019).
Jo, I., Park, Y., Lee, J. & Bae, J. A portable and spring-guided hand exoskeleton for exercising flexion/extension of the fingers. Mech. Mach. Theory 135, 176–191 (2019).
Zhang, F., Lin, L., Yang, L. & Fu, Y. Design of an Active and Passive Control System of Hand Exoskeleton for Rehabilitation. Appl. Sci. 9, 2291 (2019).
Sandoval-Gonzalez, O. et al. Design and Development of a Hand Exoskeleton Robot for Active and Passive Rehabilitation. Int. J. Adv. Robot. Syst. 13, 66 (2016).
Yurkewich, A., Hebert, D., Wang, R. H. & Mihailidis, A. Hand Extension Robot Orthosis (HERO) Glove: Development and Testing With Stroke Survivors With Severe Hand Impairment. IEEE Trans. Neural Syst. Rehabil. Eng. 27, 916–926 (2019).
NUADA. YOUR LIFE. HANDS ON. NUADA - YOUR LIFE. HANDS-ON https://nuada.pt/.
Carbonhand - Bioservo. https://www.bioservo.com/products/carbonhand.
Ironhand - Bioservo. https://www.bioservo.com/contact/customer-support/ironhand.
Shin, H. E. et al. Therapeutic Effects of Functional Electrical Stimulation on Physical Performance and Muscle Strength in Post-stroke Older Adults: A Review. Ann. Geriatr. Med. Res. 26, 16 (2022).
Deley et al. One year of training with FES has impressive beneficial effects in a 36-year-old woman with spinal cord injury. J. Spinal Cord Med. 40, 107–112 (2017).
Chou, L.-W. & Binder-Macleod, S. A. The effects of stimulation frequency and fatigue on the force–intensity relationship for human skeletal muscle. Clin. Neurophysiol. 118, 1387–1396 (2007).
Dunkelberger, N., Schearer, E. M. & O’Malley, M. K. A review of methods for achieving upper limb movement following spinal cord injury through hybrid muscle stimulation and robotic assistance. Exp. Neurol. 328, 113274 (2020).
Trotobas, C. et al. Combining Functional Electrical Stimulation (FES) to Elicit Hand Movements and a Mechanical Orthosis to Passively Maintain Wrist and Fingers Position in Individuals With Tetraplegia: A Feasibility Test. IEEE Trans. Med. Robot. Bionics 6, 1170–1179 (2024).
Wang, Y., Metcalfe, B., Zhao, Y. & Zhang, D. An Assistive System for Upper Limb Motion Combining Functional Electrical Stimulation and Robotic Exoskeleton. IEEE Trans. Med. Robot. Bionics 2, 260–268 (2020).
Cardoso, L. R. L., Bochkezanian, V., Forner-Cordero, A., Melendez-Calderon, A. & Bo, A. P. L. Soft robotics and functional electrical stimulation advances for restoring hand function in people with SCI: a narrative review, clinical guidelines and future directions. J. NeuroEngineering Rehabil. 19, 66 (2022).
Neto, A. R., Fajardo, J., Ferman, V., Fujiwara, E. & Rohmer, E. A Hybrid Control Strategy for Tendon-actuated Robotic Glove and Functional Electrical Stimulation – A Preliminary Study. in 2019 IEEE 4th International Conference on Advanced Robotics and Mechatronics (ICARM) 244–249 https://doi.org/10.1109/ICARM.2019.8834141 (2019).
Kavianirad, H., Missiroli, F., Endo, S., Masia, L. & Hirche, S. Toward Dexterous Hand Functional Movement: Wearable Hybrid Soft Exoglove-FES Study. in 2024 10th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob) 1346–1351 https://doi.org/10.1109/BioRob60516.2024.10719731 (2024).
Andreasen Struijk, L. N. S., Bentsen, B., Gaihede, M. & Lontis, R. E. Usability of the inductive tongue computer interface: Internet use, speaking, and drinking - evaluated by two users with disabilities. in 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) 1–4 https://doi.org/10.1109/EMBC40787.2023.10341041 (2023).
Hiramatsu, Y., Kimura, D., Kadota, K., Ito, T. & Kinoshita, H. Control of Precision Grip Force in Lifting and Holding of Low-Mass Objects. PLOS ONE 10, e0138506 (2015).
Mohammadi, M. et al. Eyes-Free Tongue Gesture and Tongue Joystick Control of a Five DOF Upper-Limb Exoskeleton for Severely Disabled Individuals. Front. Neurosci. 15, (2021).
Andreasen Struijk, L. N. S. An Inductive Tongue Computer Interface for Control of Computers and Assistive Devices. IEEE Trans. Biomed. Eng. 53, 2594–2597 (2006).
Mohammadi, M., Knoche, H., Gaihede, M., Bentsen, B. & Andreasen Struijk, L. N. S. A high-resolution tongue-based joystick to enable robot control for individuals with severe disabilities. in 2019 IEEE 16th International Conference on Rehabilitation Robotics (ICORR) 1043–1048 https://doi.org/10.1109/ICORR.2019.8779434 (2019).
Henderson, E., Nash, D. H. & Dempster, W. M. On the experimental testing of fine Nitinol wires for medical devices. J. Mech. Behav. Biomed. Mater. 4, 261–268 (2011).
Westerveld, A. J., Schouten, A. C., Veltink, P. H. & van der Kooij, H. Control of thumb force using surface functional electrical stimulation and muscle load sharing. J. NeuroEngineering Rehabil. 10, 104 (2013).
Struijk, L. N. S. A., Bentsen, B., Gaihede, M. & Lontis, R. Speaking Ability while Using an Inductive Tongue-Computer Interface for Individuals with Tetraplegia: Talking and Driving a Powered Wheelchair - a Case Study. in 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) 2483–2486 https://doi.org/10.1109/EMBC.2018.8512834 (2018).
Acknowledgements
This study was funded by the HyperHand project supported by a grant from Offerfonden, Denmark (Journal No. 20-610-00097), and by Aage og Johanne Louis-Hansens Fond (Grant No. 20-2B-7273). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Author information
Authors and Affiliations
Neurorehabilitation Robotics and Engineering, Center for Rehabilitation Robotics, Department of Health Science and Technology, Aalborg University, Aalborg, Denmark
Oguzhan Kirtas, Rasmus L. Kæseler, Mostafa Mohammadi & Lotte N. S. Andreasen Struijk
Faculty of Electrical Engineering, Mathematics and Computer Science, University of Twente, Enschede, The Netherlands
Peter H. Veltink
Spinal Cord Injury Centre of Western Denmark, Viborg, Denmark
Alexander L. Ammitzbøll & Ellen M. Hagen
Department of Clinical Medicine, Aarhus University, Aarhus, Denmark
Ellen M. Hagen
Department of Brain Repair and Rehabilitation, Queen Square Institute of Neurology, University College London, London, UK
Ellen M. Hagen
Authors
- Oguzhan Kirtas
- Peter H. Veltink
- Rasmus L. Kæseler
- Alexander L. Ammitzbøll
- Ellen M. Hagen
- Mostafa Mohammadi
- Lotte N. S. Andreasen Struijk
Contributions
L.N.S.A.S. and P.H.V. conceptualized the study. L.N.S.A.S. and P.H.V. secured funding and supervised the project. O.K., L.N.S.A.S., P.H.V., and M.M. designed the methodology. O.K. and M.M. developed the software. A.L.A. and E.M.H. recruited the participants. O.K., R.L.K., A.L.A., and L.N.S.A.S. investigated the study and conducted the experiments. O.K. performed formal analysis and visualization. O.K. and L.N.S.A.S. wrote the original draft, and all authors reviewed and edited the manuscript.
Corresponding author
Correspondence to Oguzhan Kirtas.
Ethics declarations
Competing interests
L.N.S.A.S. is listed as an inventor on a patent application (applicant: Aalborg University, publication number: WO/2023/046252, international application number: PCT/DK2022/050196), relevant to the non-invasive tongue-based sensor activation method described in this work. LNSAS has no ownership of the patent, which is entirely owned by Aalborg University. The other authors declare no competing financial or non-financial interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Kirtas, O., Veltink, P.H., Kæseler, R.L. et al. A tongue-controlled FES-robotic exoskeleton restores hand function in users with cervical spinal cord injury. npj Robot 4, 53 (2026). https://doi.org/10.1038/s44182-026-00117-9
Received: 13 March 2026
Accepted: 19 September 2026
Published: 08 October 2026
Version of record: 08 October 2026
DOI: https://doi.org/10.1038/s44182-026-00117-9
来源:Nature 健康护理主题 · nature.com