A multidimensional evaluation of the Atalante exoskeleton as an assistive locomotion device for individuals with complete paraplegia: a prospective longitudinal study in a controlled environment Authors: Eric Pantera, Nicolas Reneaud, Claude Bouscavert, Didier Pradon Published: 2026 ABSTRACT AbstractBackgroundRobotic exoskeletons are increasingly used in neurorehabilitation, yet evidence remains limited regarding their role as assistive locomotion devices within a compensatory rehabilitation framework, particularly in individuals with complete spinal cord injury. Multidimensional evaluations integrating safety, usability, and patient-reported experience are still scarce.ObjectiveTo evaluate the Atalante self-balancing exoskeleton as an assistive locomotion device in individuals with complete paraplegia through a multidimensional analysis conducted in a controlled clinical environment.MethodsThis prospective, monocentric, longitudinal study included adults with chronic complete thoracic spinal cord injury (American Spinal Injury Association Impairment Scale grade A, neurological level T6–T12). Participants underwent an intensive 6-week program comprising 60 supervised sessions of exoskeleton-assisted training including 40 m walking per session. Outcomes were assessed across three complementary dimensions: (1) safety (medical adverse events, safety-related technical incidents, near-falls); (2) techno-functional usability (donning/doffing performance, assistance levels, cumulative walking exposure); and (3) acceptability (patient-reported comfort and perceived safety using session-based visual analogue scales). Physiological responses were evaluated using exoskeleton-assisted 6-minute walk tests.ResultsThirteen participants were enrolled; eleven completed the full protocol. One serious adverse event occurred, a tibial plateau fracture in a participant with severe osteoporosis (hip T-score − 3.8), leading to study withdrawal. No other serious medical adverse events were observed. Across 651 sessions, four near-fall events were recorded (0.61% of sessions and 0.16% per 10 m walked). Minor adverse events were transient, and no symptomatic orthostatic hypotension was reported. Oxygen consumption, walking speed, and cost of transport (p > 0.05) remained stable between mid- and end-program assessments, while perceived exertion decreased significantly (p = 0.003). Donning time decreased from 790 [475–966] s during the first session to 223 [159–259] s during the final sessions, and doffing time from 242 [152–275] s to 55 [50–69] s, with familiarization plateau points observed after 14 and 18 sessions, respectively. Harness comfort improved over time, whereas perceived safety remained globally stable, except for transient decreases following fall-prone gait events or technical disengagements.ConclusionsUnder supervised clinical conditions, the Atalante exoskeleton demonstrated acceptable safety and feasibility as an assistive locomotion device for individuals with complete paraplegia, enabling hands-free upright indoor walking with low metabolic demand. Severe osteoporosis emerged as a critical risk factor, supporting systematic bone densitometry and cautious exclusion criteria. These findings support the integration of self-balancing exoskeletons into compensatory rehabilitation programs and justify further studies in ecological settings to assess real-world autonomy and participation outcomes.Trial registration ClinicalTrials.gov Identifier: NCT06941896 (retrospective registration)Keywords: Exoskeleton, Paraplegia, Assistive locomotion, Rehabilitation, Safety, Balance control Introduction Restorative and compensatory rehabilitation are closely related but distinct concepts, each corresponding to complementary therapeutic objectives in patient care. Restorative rehabilitation refers to interventions aimed at inducing measurable changes through targeted work on altered body functions [1], relying on the assumption that functional improvement is possible. In contrast, compensatory rehabilitation focuses on strategies that support the execution of activities and participation despite persistent impairments [2–4]. Whereas restorative approaches primarily target improvements in body functions, compensatory rehabilitation seeks to optimize autonomy and quality of life in the presence of enduring limitations [3, 4]. Within this framework, technological innovation plays differentiated roles: in restorative rehabilitation, devices are designed to stimulate impaired body functions [5, 6], whereas in compensatory rehabilitation, technologies act as assistive solutions supporting activity and participation [6, 7]. These two approaches correspond to distinct therapeutic logics, underscoring the importance of clearly aligning technology use with either restorative or compensatory objectives [7]. In the context of locomotion, individuals with complete spinal cord injury (SCI), whether paraplegia or tetraplegia, provide a clinically relevant model to illustrate this distinction [8]. In complete SCI, the potential for neurological restoration of voluntary lower-limb function is extremely limited, which places compensatory mobility strategies at the core of long-term rehabilitation planning [9]. Within this context, robotic exoskeletons have emerged as a major technological innovation, supporting both restoration-oriented and compensatory objectives [7–9]. From a restorative rehabilitation perspective, several studies suggest that exoskeleton-assisted walking may enhance neuromuscular activation [5, 9], improve cardiovascular conditioning [6], and reduce complications associated with prolonged immobility [10], although results remain heterogeneous and often limited by short intervention durations or small sample sizes. From a compensatory rehabilitation perspective, however, multiple studies [5, 9, 11] emphasize that exoskeletons primarily function as assistive locomotion devices compensating for absent motor function, rather than as tools for neuromotor recovery. In practice, their current use mainly enables upright indoor mobility in controlled environments rather than fully integrated community ambulation [4]. Accordingly, exoskeletons should be considered personal assistive technologies supporting upright functional mobility and interaction with the environment, rather than devices aimed at restoring autonomous community walking. To address some limitations of existing systems, several humanoid robotics developers have designed exoskeletons combining self-balancing capabilities with assisted gait [12, 13]. Among them, the Atalante exoskeleton developed by Wandercraft [14] enables hands-free, dynamically self-stabilized walking without upper-limb support. This configuration aligns with a compensatory rehabilitation paradigm centered on upright mobility and interaction with the environment. Nevertheless, available evidence remains limited and largely derived from evaluations conducted in controlled indoor settings, and most studies focus on isolated outcomes without adopting a genuinely multidimensional rehabilitation perspective integrating technical, clinical, and subjective components of device use [4]. Although the real-world relevance of such technologies ultimately requires validation in ecological conditions, a necessary preliminary step is to examine their use and effects in controlled environments within the target population [15]. Accordingly, the primary aim of the present study was to evaluate the Atalante exoskeleton through a multidimensional analysis in individuals with complete paraplegia. The protocol was structured around three complementary dimensions: (1) safety, assessed through biomedical monitoring; (2) usability, evaluated through techno-functional performance; and (3) acceptability, examined through users’ perceived confidence in the device. Methods Atalante exoskeleton The Atalante exoskeleton (version 1.0, 2018; Wandercraft, Paris, France) is a self-balancing, crutch-less lower-limb exoskeleton designed to enable upright locomotion without additional external assistive devices. The mechanical architecture includes 12 actuated degrees of freedom: three at the hip (flexion/extension, abduction/adduction, internal/external rotation), one at the knee (flexion/extension), and two at the ankle (plantar-/dorsiflexion and rotation along Henke’s axis), symmetrically replicated on each lower limb [12–14]. A passive forefoot joint allows natural foot rollover during stance. Motor control relies on a hybrid continuous/discrete-time architecture inspired by orbital dynamic models. Walking trajectories are precomputed using nonlinear optimization based on direct collocation and tracked in real time using a hybrid zero dynamics controller stabilized by state feedback [14]. This control strategy stabilizes the human–exoskeleton system by enforcing virtual kinematic constraints and regulating state deviations through feedback control, thereby constraining gait to a stable periodic pattern (limit cycle) and ensuring dynamic postural stability throughout both stance and swing phases, as previously described in detail by Grizzle et al. [16]. User intent detection is achieved through sensor fusion combining inertial measurement units (IMUs), pressure sensors, and joint encoders. Several operational modes are implemented (sit-to-stand, balance, walk, stop), selected via a user interface that can be operated by either the therapist or the patient, depending on the session’s objectives. In the event of a detected anomaly, safety modes trigger an immediate return to a stabilized standing posture. Load-bearing zones (pelvis, thighs, knees, lower legs) consist of padded rigid shells secured by adjustable straps to ensure fixation and comfort. A ceiling-mounted overhead rail and harness system provided passive fall arrest; harness engagement occurred only in the event of balance loss and did not modify the exoskeleton’s control behavior (Fig. 1). Atalante Exoskeleton: A Atalante exoskeleton (v1.0). B Participant wearing the device during the clinical protocol Study design This prospective, longitudinal, monocentric, single-group study evaluated the use of the Atalante exoskeleton as an assistive locomotion device within a structured compensatory rehabilitation program focused on upright indoor mobility in a controlled clinical environment. The study was approved by the local ethics committee (RCB number: 2016-A01221-50) and conducted in accordance with the Declaration of Helsinki (Fortaleza revision, 2013). All participants provided written informed consent prior to inclusion. Participants Adults with chronic complete thoracic spinal cord injury (SCI) were consecutively screened between September 2018 and December 2019. Inclusion criteria were: complete SCI classified as American Spinal Injury Association Impairment Scale (AIS) Grade A according to the International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI) [17] ; a neurological level of injury (NLI) between T6 and T12, selected to allow effective triggering of the exoskeleton through controlled forward trunk inclination and reliable detection of user intent via IMU signals; time since injury ≥ 6 months; and anthropometric characteristics compatible with the exoskeleton, including a body height between 1.50 and 1.90 m, a thigh length of approximately 380–460 mm, a knee-to-ground distance compatible with device adjustment, and a body mass ≤ 90 kg. Absence of symptomatic orthostatic intolerance was required and defined as the absence of dizziness, presyncope, or a decrease in systolic blood pressure ≥ 20 mmHg during a standardized upright standing trial performed in a verticalization device (standing frame) maintained for 5 min, in accordance with established clinical recommendations for orthostatic tolerance assessment in individuals with SCI [18]. Exclusion criteria were: lower-limb spasticity > 3 on the Modified Ashworth Scale, assessed separately at the hip, knee, and ankle joints [14] ; bilateral lower-limb range-of-motion limitations incompatible with exoskeleton trajectories; spinal instability; severe psychiatric or cognitive disorders; recent dermatological conditions at contact areas; pregnancy or breastfeeding; cardiopulmonary contraindications to exercise; a history of lower-limb fragility fracture within the previous two years; or the presence of five or more fracture risk factors during exoskeleton-assisted walking, as defined by Craven et al. [19]. Baseline neurological assessment included AIS Grade, NLI, Total Motor Score, Total Sensory Score, and sensory Zone of Partial Preservation (ZPP), assessed according to ISNCSCI standards [17]. Rehabilitation and evaluation procedure The intervention was delivered over six weeks, with two 45-minute sessions per day, five days per week, for a total of 60 planned sessions per participant. Sessions were supervised by an experienced physiotherapist or an adapted physical activity professional trained in exoskeleton-assisted mobility. Each session followed a standardized sequence of five tasks: donning, sit-to-stand, self-balancing, straight-line walking, and doffing. Donning was performed from the participant’s wheelchair positioned in front of the exoskeleton. With verbal instructions and, when necessary, physical assistance, participants were transferred and secured using the pelvic and lower-limb straps. Donning time was defined as the interval from the initiation of the transfer from the wheelchair to the completion of fastening of the last exoskeleton strap. The level of assistance was recorded for each session as: none (verbal supervision only), partial (intermittent physical assistance), or full (continuous physical assistance). Sit-to-stand was performed using a dedicated exoskeleton mode triggered by forward trunk inclination detected by IMUs. After a familiarization phase, participants executed this task without physical assistance; the supervising professional remained in close standby position. Self-balancing was assessed in active balance mode, during which the exoskeleton maintained upright posture without upper-limb support. Participants performed voluntary reaching tasks toward targets placed anteriorly and laterally by the supervising professional (approximately five repetitions per direction), inducing self-generated perturbations without external pushes. Straight-line walking consisted of four 10-meter bouts (total distance: 40 m) along an indoor walkway under an overhead rail system. Walking was performed without crutches or walkers. Due to software limitations of Atalante v1.0, turning was not autonomous; a therapist-assisted 180° turn was performed between bouts after gait termination. Doffing time was defined as the interval from the initiation of the unstrapping process to the completion of the transfer back into the wheelchair, using the same assistance scale as for donning. Stand-to-sit was systematically performed for safety but was not analyzed as an independent outcome. Safety (safety dimension) Medical adverse events (AEs) were defined as any new or worsening clinical sign or symptom occurring during or within 24 h after a session, regardless of causality. These included, but were not limited to, fractures, joint pain or swelling, skin abrasions or pressure injuries, orthostatic hypotension, and other musculoskeletal or systemic complaints. All AEs were classified as device-related or non–device-related, and as serious or non-serious. Safety-related technical events included trajectory deviations and perturbed gait (characterized by shuffling episodes, abrupt foot strikes, marked lateral instability, and tripping), fall-prone gait (defined as unintended backward displacement of the user–exoskeleton system during walking), safety disengagements (automatic transition to a safety mode with return to a stabilized posture to prevent a fall), and near-fall events (harness activations providing effective weight support). All events were documented systematically after each session in a dedicated logbook. Physiological assessment Physiological cost was assessed using a 6-minute walk test (6MWT) performed with the exoskeleton during session 21 (beginning of week 3) and session 60 (end of week 6). Oxygen uptake was measured using a portable metabolic system (K5, COSMED, Rome, Italy), with breath-by-breath data averaged over the final minute of the test, corresponding to steady-state conditions. Walking speed was derived from the total distance covered, and cost of transport (CoT) was calculated as oxygen consumption divided by walking speed [20]. Perceived exertion was assessed using the Borg scale. Usability / techno-functional dimension Total walking distance per session and the total number of completed sessions were extracted from the standardized session reports documented by the clinical team. Cumulative exposure was defined as the sum of all walking distances across the protocol. Donning and doffing times were measured at every session with a stopwatch, and assistance levels were recorded systematically. Technical incidents were documented in parallel. These data were used to characterize learning curves and familiarization. User satisfaction was assessed at the end of week 6 using the Quebec User Evaluation of Satisfaction with Assistive Technology (QUEST 2.0) [21]. Patient-reported outcomes (acceptability) Perceived comfort and safety were assessed after each session using visual analogue scales (VAS). All VAS ranged from 0 (minimal comfort or safety) to 10 (maximal comfort or safety). Subdimensions included comfort during donning/doffing and perceived safety during standing and walking. VAS scores were used to analyze longitudinal changes and their relationship with technical incidents and adverse events. VAS subdimensions included overall comfort, strap and thoracic harness comfort, and perceived safety during sitting, sit-to-stand, stabilization, balance, and walking. In this context, “stabilization” referred to perceived safety while maintaining a stable upright standing posture, whereas “balance” referred to perceived safety during active balance mode involving voluntary perturbations (e.g., forward and lateral reaching tasks). Statistical analyses Quantitative data were expressed as mean ± standard deviation or as median with interquartile range, depending on data distribution assessed using the Shapiro–Wilk test and inspection of skewness. Within-subject pre-/post-training comparisons were performed using paired Student’s t-tests for normally distributed variables and Wilcoxon signed-rank tests for non-parametric data. The statistical significance threshold was set at p < 0.05 for all analyses. Effect sizes (Cohen’s d and rank-biserial correlation) with their 95% confidence intervals were reported for each comparison. The familiarization plateau for donning and doffing was defined as the point at which further reductions in task duration became marginal across successive sessions. Performance evolution curves were analyzed over the entire training period to identify the inflection point beyond which additional improvements were minimal. To reduce inter-session variability, time-series data were smoothed using a moving average with a 3-session window. The inflection point was defined as the moment when the smoothed execution time reached 90% of the stable execution time observed at the end of the protocol (the final 10 sessions). Inflection point detection was based under the assumption that early sessions primarily reflect rapid learning effects, whereas later sessions better represent stable, reinforced task execution under standardized supervision conditions. Results Study population and protocol completion A total of 13 individuals initiated the study and constituted the safety population. Statistical analyses were conducted on data from 11 participants who completed the full protocol and therefore formed the per-protocol population. Two participants discontinued participation: Subject 3 was withdrawn following a serious adverse event (tibial plateau fracture) that occurred during an exoskeleton-assisted walking session at session 7, and subject 13 withdrew voluntarily after 22 sessions due to a perceived lack of functional benefit. Participant characteristics Baseline demographic, anthropometric, and neurological characteristics of the 13 enrolled participants are summarized in Table 1. All participants presented with chronic complete paraplegia, classified as AIS Grade A according to ISNCSCI [17] with neurological levels ranging from T6 to T12. Mean age was 45.4 ± 11.4 years, and mean body mass index (BMI) was 24.8 ± 4.4 kg·m⁻². Median time since injury was 7.3 years [1.8–28.7]. Total Motor Score, Total Sensory Score, NLI, and sensory Zone of Partial Preservation (ZPP) are reported in Table 1. Table 1Participant characteristicsIDSexAge(years)Height (m)Weight (kg)Time since injury (years)Total motor score (/100)Total sensory score(/224)AIS gradeNLIZPPTWD(m)Sessions (n)1M341.75652.750104AT6T62320602M321.72623.550128AT9T102400603M581.708028.750136AT10T1126074M571.675715.850120AT8T82280585F411.62509.650112AT7T92200606F331.55674.250152AT12T122320607M321.80605.650112AT7T92320608M361.76647.350120AT8T82360609M561.859014.150112AT7T822806010M571.789017.350104AT6T824006011M541.62872.650152AT12T1222805712M371.73861.85090AT7T922405613F571.68705.750104AT6T882022AIS = ASIA Impairment ScaleNLI = Neurological Level of Injury; ZPP = Zone of Partial Preservation; TWD = total walking distance Participant characteristics AIS = ASIA Impairment Scale NLI = Neurological Level of Injury; ZPP = Zone of Partial Preservation; TWD = total walking distance Safety dimension Medical adverse events Across all sessions, a total of 36 medical AEs were recorded on 33 sessions. These included 21 episodes of pain, considered device-related, as well as 13 episodes of spasticity, one case of dermabrasion, and one episode of tachycardia, considered related to the training procedure. Event frequencies were normalized per session for each participant and are summarized in Fig. 2 using a radar plot representation, providing an overview of the relative distribution of AE categories across the entire protocol. Across the entire study period, one serious medical adverse event was observed: a tibial plateau fracture in Subject 3, occurring during an exoskeleton-assisted walking task, which led to immediate withdrawal from the study. No other serious medical AEs were reported among participants who completed the protocol. Non-serious medical AEs were predominantly transient and resolved spontaneously or after temporary adjustment of training intensity. No episodes of syncope or clinically significant orthostatic hypotension were observed. Distribution of safety-related medical and technical events during exoskeleton-assisted training Safety-related technical events and near-falls Across all sessions, a total of 516 safety-related technical events were recorded on 380 sessions, including 219 trajectory deviations, 165 perturbed gait patterns, 28 fall-prone gait events, 104 safety disengagements, and 4 near-fall events. Event frequencies were normalized per session for each participant and are summarized in Fig. 2. Overall, this corresponded to 4 near-fall events across 651 sessions (0.61% of sessions). When referenced to the total walking exposure, this corresponded to approximately 0.16% per 10 m of walking. Analysis of temporal distribution showed that technical events occurred predominantly during early training sessions, with a progressive reduction over time, whereas near-fall events remained rare and were not clustered in late training phases. Radar plot illustrating the relative frequency of medical adverse events and safety-related technical incidents across all sessions, grouped by event category. Values are normalized per session and expressed as occurrence percentages (median with interquartile range, Q1–Q3). Usability / techno-functional dimension Donning and doffing performance Donning and doffing times decreased substantially over the training period. Median donning time decreased from 790 [475–966] seconds during the first session to 223 [159–259] seconds during the final sessions. Median doffing time decreased from 242 [152–275] seconds to 55 [50–69] seconds over the same period. Familiarization analysis based on smoothed time-series curves, averaged across the 11 participants who completed the protocol, identified median plateau points at 14 [12–18] sessions for donning and 18 [11–22] sessions for doffing. After these points, further reductions in task duration were marginal. Mean smoothed learning curves for donning and doffing are illustrated in Fig. 3. Donning and doffing learning curves Assistance level The level of therapist assistance required for donning and doffing was recorded for every session. Over the course of training, the proportion of sessions performed without physical assistance increased progressively for both tasks, reflecting a clear familiarization effect. The number of sessions required to reach independence for donning and doffing across participants is illustrated in Fig. 3. Straight-line walking was performed hands-free in all sessions, with the therapist remaining in close standby without physical contact. Physical contact occurred only in occasional situations where safety-related events arose (e.g., trajectory deviations or gait perturbations), requiring brief manual stabilization to ensure participant safety. Due to device limitations (Atalante v1.0), turning required therapist assistance in all sessions. The curves illustrate the progressive reduction in median donning (top) and doffing (bottom) times across training sessions. The shaded areas represent the interquartile range (Q1–Q3), and the dashed vertical line indicates the median number of sessions required to achieve task independence. The purple boxplots represent the distribution across participants of the number of sessions required to reach full independence in donning and doffing. Walking exposure The total walking distance for each participant is presented in Table 1. Cumulative exposure to the exoskeleton throughout the study amounted to a total of 25,400 m across all completed sessions. Physiological dimension Physiological parameters were assessed during exoskeleton-assisted 6-minute walk tests (6MWT) conducted at session 21 (mid-program) and session 60 (end of training). No significant pre-/post-training differences were observed for oxygen consumption (VO₂), walking speed, or cost of transport (CoT). Mean VO₂ was 7.96 ± 2.67 mL·kg⁻¹·min⁻¹ at session 21 and 7.63 ± 2.30 mL·kg⁻¹·min⁻¹ at session 60 (p = 0.619). Walking speed was 0.16 ± 0.03 m·s⁻¹ versus 0.15 ± 0.03 m·s⁻¹ (p = 0.098), and CoT was 0.87 ± 0.37 versus 0.86 ± 0.29 mL·kg⁻¹·m⁻¹ (p = 0.890). In contrast, perceived exertion during the 6MWT, assessed using the Borg scale, decreased significantly from 2.6 ± 0.8 to 1.1 ± 0.8 (p = 0.003), indicating reduced subjective effort for comparable walking performance (Table 3). Table 3Pre- and post-protocol clinical outcomes after four weeks of exoskeleton-assisted walkingVariablesPre (SD)Post (SD)p-valueEffect size (95% CI)VO₂ (mL·kg⁻¹·min⁻¹)7.96 (2.67)7.63 (2.30)0.6190.155 (-0.444–0.746)V (m·s⁻¹)0.16 (0.03)0.15 (0.03)0.0980.550 (-0.099–1.175)COT (mL·kg⁻¹·m⁻¹)0.87 (0.37)0.86 (0.29)0.8900.043 (-0.549–0.633)RPE (Borg scale)2.64 (0.81)1.09 (0.83) 0.003 1.000 (1.000–1.000)* VO₂: oxygen consumption; V: walking speed; COT: cost of transport; RPE: perceived exertion scale. Values are expressed as mean (SD). The last column reports the effect size (Cohen’s d) with the 95% confidence interval (CI). *For RPE, the rank-biserial correlation was used due to the non-normal distribution of the Borg scale data; the value of 1.000 reflects a consistent decrease across all participants Pre- and post-protocol clinical outcomes after four weeks of exoskeleton-assisted walking VO₂: oxygen consumption; V: walking speed; COT: cost of transport; RPE: perceived exertion scale. Values are expressed as mean (SD). The last column reports the effect size (Cohen’s d) with the 95% confidence interval (CI). *For RPE, the rank-biserial correlation was used due to the non-normal distribution of the Borg scale data; the value of 1.000 reflects a consistent decrease across all participants Patient-reported outcomes (acceptability) Patient-reported comfort and perceived safety were assessed after each session using visual analogue scales across eight predefined subdimensions. Among these, thoracic harness comfort showed a statistically significant improvement between the beginning and the end of the program, increasing from 8.0 [5.5–8.0] to 8.0 [7.5–8.5] (p = 0.035). All other comfort and safety-related VAS scores remained stable over time. Pre- and post-training distributions for each subdimension are displayed in Fig. 4. VAS pre/post (scale 0–10). Six-week training protocol assessing perceived safety and comfort. Statistically significant pre/post difference: * p-value < 0.05, “+” symbols represent outliers in the boxplot Impact of technical incidents on perceived safety The acute impact of safety-related technical events on patient-reported outcomes was evaluated by comparing VAS scores before and after sessions involving incidents. Among these events, only four corresponded to near-fall situations. Following fall-prone gait, perceived safety during walking decreased by − 0.59 ± 1.21, and perceived safety during sit-to-stand transitions decreased by − 0.33 ± 1.36. Exoskeleton disengagement was associated with a reduction in perceived walking safety (− 0.71 ± 2.01), while perturbed gait was associated with a smaller decrease (-0.31 ± 1.36). Changes in other comfort and safety subdimensions were minimal and lacked a consistent pattern. Detailed results are presented in Table 4. Table 4Effect of safety-related technical events on visual analogue scale (VAS) scores for comfort and safetyVASTrajectory deviation(219)Perturbedgait(165)Fall-prone gait(28)Disengagement(104)Diff. (SD)Diff. (SD)Diff. (SD)Diff. (SD)Overall comfort0.05 (1.03)−0.05 (0.96)−0.26 (0.59)−0.20 (0.93)Strap comfort0.01 (0.74)−0.04 (0.62)0.00 (0.28)0.01 (0.74)Harness comfort0.00 (0.37)−0.01 (0.59)−0.04 (0.52)−0.04 (0.56)Safety while sitting0.00 (0.35)−0.05 (0.50)−0.04 (0.19)−0.11 (0.56)Safety during sit-to-stand transition0.00 (0.52)−0.10 (1.01)−0.33 (1.36)−0.23 (1.19)Safety during stabilization0.01 (0.45)−0.05 (0.47)−0.04 (0.19)−0.08 (0.48)Safety related to balance0.00 (0.39)−0.04 (0.47)0.00 (0.28)−0.04 (0.46)Safety while walking0.07 (1.31)−0.31 (1.36)−0.59 (1.21)−0.71 (2.01)Values represent the mean difference (SD) between VAS scores recorded before and after sessions involving technical incidents. The number of incident occurrences is indicated in parentheses Effect of safety-related technical events on visual analogue scale (VAS) scores for comfort and safety Values represent the mean difference (SD) between VAS scores recorded before and after sessions involving technical incidents. The number of incident occurrences is indicated in parentheses Discussion This study evaluated the Atalante lower-limb exoskeleton primarily as an assistive locomotion device in a compensatory rehabilitation perspective, rather than as a neurorestorative intervention aiming at neurological recovery in individuals with complete paraplegia. While several studies have already explored exoskeleton-assisted walking in persons with thoracic AIS A spinal cord injury with compensatory objectives [7–9], the added value of the present work lies in: (i) the high-dose exposure to a self-balancing system (60 sessions over six weeks, twice daily), (ii) a multidimensional framework combining safety, techno-functional usability, and patient-reported outcomes, (iii) the session-by-session integration of perceived safety/comfort with documented safety-related technical events, and (iv) the characterization of familiarization dynamics for donning and doffing with systematic recording of assistance levels. Eleven participants completed the six-week protocol, allowing assessment of feasibility and tolerability of repeated exposure under standardized clinical supervision. The protocol intentionally focused on prerequisite tasks for exoskeleton use, donning, sit-to-stand, self-balancing, straight-line walking, and doffing, rather than attempting to reproduce the full complexity of ecological home or community environments. This controlled setting was chosen as an intermediate step prior to more ecological deployment, similarly to other functional simulation approaches sometimes used to prepare discharge planning [22]. Safety dimension In a compensatory rehabilitation paradigm, safety is central, as the device is used to compensate for permanent motor deficits and is therefore exposed to repeated use over time. Accordingly, medical adverse events and safety-related technical incidents were systematically recorded throughout the protocol. Several studies have reported adverse events such as orthostatic hypotension [23], pain, and skin irritation [7] during exoskeleton-assisted walking, sometimes requiring protocol adjustments or session discontinuation, highlighting their potential impact on feasibility and adherence. In the present study, cutaneous and hemodynamic tolerance was overall favorable, with mainly mild, transient, non-serious events and no clinically meaningful episodes of symptomatic orthostatic hypotension during repeated verticalization. These observations are consistent with studies reporting low rates of clinically significant intolerance when exoskeleton use is embedded within a structured clinical framework, including prior screening for contraindications and close hemodynamic and functional monitoring during training sessions [7, 9]. A major safety finding was the occurrence of one serious adverse event: a tibial plateau fracture in a participant with a long-standing spinal cord injury (28.7 years post-trauma), withdrawn at session 7. Importantly, this event prompted retrospective bone densitometry, which revealed severe osteoporosis with a hip T-score of − 3.8. Retrospective bone screening was then performed for all other participants, and no other participant had a hip T-score < − 3.5. At the time the study was conducted, there were no widely established clinical recommendations defining a specific hip T-score threshold for the safe use of powered exoskeletons in individuals with spinal cord injury, and bone health screening was therefore performed according to routine clinical practice and commonly applied risk-factor–based strategies [19]. This case suggests that commonly applied screening strategies [19] may remain insufficiently conservative in individuals with long-standing spinal cord injury, even in the absence of recent fracture history or cumulative risk thresholds. In line with existing clinical recommendations and trial data on bone health in SCI populations [19, 24], but extending beyond currently proposed risk-based algorithms, our findings support strengthening pre-use bone screening by advocating systematic bone densitometry prior to exoskeleton-assisted walking, including in individuals who do not meet conventional risk-factor criteria. Based on this serious adverse event and the retrospective densitometry findings, we further propose that exoskeleton-assisted walking should be contraindicated in individuals with spinal cord injury presenting a hip T-score below − 3.5. This hypothesis, that severe hip osteoporosis is associated with an unacceptably high fracture risk during exoskeleton-assisted walking, requires confirmation in larger, dedicated cohorts. Notably, recent clinical guidance [25] on patient selection for powered exoskeleton use has also suggested similar thresholds aimed at minimizing fracture risk, further supporting the relevance of this proposed safety limit. Beyond medical adverse events, safety-related technical incidents were frequent, including trajectory deviations, perturbed gait events, fall-prone gait events, near-fall events and safety disengagements. Importantly, all sessions were conducted under close supervision and with a ceiling-mounted rail and harness system providing fall arrest. Within this controlled safety configuration, near-fall events, defined as effective harness weight-support activations, were rare, with 4 near-fall events recorded over 651 sessions (0.61%), corresponding to 0.16% per 10 m walked based on the total cumulative distance covered during the study (25,400 m). This low incidence suggests that, despite frequent minor technical perturbations, the actual risk of loss of balance requiring external support remained limited during supervised use. The overhead harness system was essential to ensure participant safety during intensive and repeated exposure and to allow progressive familiarization with the device. However, this protective setup also limits direct extrapolation to unsupervised real-world conditions, where such external fall-prevention systems are not available. Consequently, safety interpretations should be restricted to supervised clinical contexts, and near-fall events should be interpreted as proxy indicators of instability risk, rather than as estimates of real-world fall incidence. Finally, from the perspective of exercise safety and physiological load, robotic gait systems have generally been described as inducing low-to-moderate cardiopulmonary demands in SCI populations [26, 27]. In our protocol, VO₂, walking speed, and cost of transport remained stable between mid-program and end-program assessments, while perceived exertion (Borg) decreased significantly. This pattern suggests that intensive repeated use did not induce progressive metabolic overload and that reduced perceived effort likely reflects familiarization and improved task confidence rather than major physiological adaptation. Comparisons with other devices should remain cautious, as energy cost is strongly influenced by neurological level, trunk control requirements, walking speed, device architecture, and experimental conditions. For instance, higher energy costs reported in small cohorts using other systems may be partly explained by higher-level injuries (e.g., up to T1) and greater trunk instability [28], while other studies report substantially faster walking speeds than achieved in our protocol [20]. In this context, differences in device architecture may further contribute to this variability: Exoskeletons requiring external walking aids may place additional demands on the upper body, potentially increasing metabolic expenditure. Preliminary studies indicate that crutch-related loading is tied to upper-limb effort and may be reduced through optimized control strategies [29]. Likewise, gait performed with external arm support has been associated with higher oxygen consumption compared with unsupported conditions [30]. These observations suggest that upper-limb involvement can influence metabolic demand, although current evidence remains limited and largely indirect. In this context, the hands-free design of the Atalante exoskeleton may help moderate metabolic demands, although such comparisons remain indirect and should be interpreted cautiously. Techno-functional dimension In the context of assistive compensatory rehabilitation, techno-functional aspects play a central role in clinical feasibility and routine integration of robotic devices, with donning and doffing times recognized as key determinants of clinical applicability [4]. Wright et al. [7] reported a mean combined donning–doffing time of 6 min 50 s for the ABLE exoskeleton. However, direct comparison between studies is limited due to differences in participant characteristics, particularly the inclusion of individuals with tetraplegia who may require greater assistance for donning and doffing due to impaired hand function and trunk stability. In the present study, the final median donning time was 3 min 43 s [2 min 39 s–4 min 19 s] and a median doffing time of 55 s [50–69 s], durations that are compatible with routine clinical use of the device. Interpretation of donning and doffing times across studies must also consider the level of assistance required. In the present study, assistance was systematically quantified alongside time measures. Reductions in donning and doffing times occurred in parallel with decreasing physical assistance, allowing interpretation in terms of functional autonomy. At comparable exposure levels (within the first 10–12 sessions), many participants still required supervision or minimal assistance, consistent with Wright et al. [7] and Maggioni et al. [6]. Familiarization timelines vary widely, with Park et al. [5] reporting autonomy around the sixth session and Maggioni et al. [6] after 7–10 sessions. In the present study, median plateau points were identified at 14 [12–18] sessions for donning and 18 [11–22] sessions for doffing, reflecting the longer training exposure compared with previous protocols. Taken together, these findings indicate that familiarization timelines depend strongly on training duration and assistance profiling. As all participants were AIS Grade A (ISNCSCI) [17], differences across studies should be interpreted primarily in light of exposure length and assistance conditions rather than intrinsic device performance. Beyond timing metrics, cumulative exposure provides essential context for interpreting safety and feasibility outcomes. In the standardized protocol, straight-line walking was performed over approximately 40 m per session. Although this value may appear modest, the cumulative exposure across all participants reached approximately 25.4 km of exoskeleton-assisted walking performed under controlled and supervised conditions. Based on reported training parameters in previous exoskeleton studies, estimated cumulative walking exposure ranges from approximately 11.9 km [9] to 56 km [31] and more than 80 km in larger multicenter trials [8]. Therefore, despite shorter walking bouts per session, the present study still represents a meaningful overall exposure to exoskeleton-assisted walking. Anchoring adverse events and safety-related technical incidents to this cumulative exposure strengthens their interpretability, allowing incident rates to be contextualized relative to the actual dose of exoskeleton use rather than isolated session-based observations. Technical incidents (safety disengagements, trajectory deviations, fall-prone gaits) represent an additional techno-functional constraint. In supervised conditions, such incidents may be clinically manageable, but they can influence both therapist confidence and user acceptance over repeated exposure. Accordingly, incident logs were cross-referenced with session-by-session perceived safety and comfort measures to quantify their acute subjective impact. These results support the clinical relevance of combining objective technical logs with patient-reported outcomes to better capture usability constraints that are not reflected by performance metrics alone. Acceptability and perceived confidence dimension In compensatory assistive technologies, acceptability depends not only on technical feasibility but also on perceived comfort, safety, and confidence, which influence adherence and long-term engagement. Qualitative work emphasizes the importance of structured support and realistic expectation setting early in training [11, 32]. In our study, overall satisfaction assessed by QUEST 2.0 was moderate (mean 3.69/5). This value appears slightly lower than that reported with other systems in supervised contexts [7], which may reflect the use of an early clinical version of the device (Atalante v1.0), the high intensity of exposure, and the occurrence of technical interruptions over a prolonged program. In addition, recent discussions highlight limitations of QUEST 2.0 in fully capturing user experience dimensions such as perceived safety and comfort [33], supporting the complementary use of session-based VAS. Indeed, VAS outcomes provided a dynamic perspective on acceptability. A significant improvement was observed specifically for harness comfort over the course of the protocol, particularly during familiarization with the thoracic harness. In contrast, safety-related VAS scores remained globally stable over time, except following isolated technical incidents. Notably, fall-prone gait events were associated with transient decreases in perceived safety, with mean reductions of − 0.59 points for safety during walking and − 0.33 points for safety during sit-to-stand transitions. Changes in other comfort dimensions remained modest and non-significant. However, these incidents were accompanied by transient reductions in perceived safety during walking and sit-to-stand transitions, suggesting that software- or system-related interruptions may negatively influence the perceived user experience, even in the absence of lasting clinical consequences. This finding is consistent with observations by Kozlowski et al. [31], who reported that technical interruptions may induce frustration and reduce motivation independently of perceived functional benefit. Together, these results highlight the importance of jointly assessing technical performance and patient-reported experience over time, as transient perturbations may meaningfully affect perceived safety and comfort despite stable objective performance under supervised conditions. Limitations Several limitations must be acknowledged. First, the study was conducted in a controlled indoor environment and therefore does not allow direct conclusions regarding autonomy or participation in community or home settings. The protocol targeted prerequisite tasks and simulated functional sequences required for device use, but it did not assess ecological outcomes such as unsupervised transfers, household navigation, outdoor walking, or social participation. Second, the overhead rail and harness system was essential to ensure safety during intensive exposure and likely contributed to higher perceived safety and to the prevention of falls that might otherwise have occurred. Therefore, perceived safety outcomes and near-fall metrics must be interpreted within the supervised protected context, and cannot be generalized to autonomous real-world use scenarios. Third, the device version (Atalante v1.0) imposed limitations such as non-autonomous turning, requiring therapist assistance, which impacts translatability to independent daily life mobility. Consequently, standardized functional assessment frameworks such as the one proposed by van Dijsseldonk et al. [34] could not be applied in the present study, as they require turning and variable-task modules that were not available in this device version. Finally, although therapist assistance and supervision were integral to the protocol and were recorded, the present design was not intended to formally model the progressive reduction of supervision as an outcome. Future studies should incorporate ecologically anchored designs, progressively reducing external constraints and documenting supervision requirements over time, in order to better characterize real-world adoption and the assistive potential of self-balancing exoskeletons. Conclusion This study provides a multidimensional evaluation of the Atalante exoskeleton as an assistive locomotion device within a compensatory rehabilitation framework for individuals with complete paraplegia, using an intensive 6-week protocol in a controlled clinical environment. Under supervised conditions and with an overhead fall-prevention system, repeated use was associated with a favorable safety profile, supported by a cumulative exposure to the device corresponding to a total walking distance of 25,400 m and a very low incidence of near-fall events (4 near-fall events across 651 sessions, corresponding to 0.61% of sessions and 0.16% per 10 m walked), with systematic recording of medical adverse events and technical safety events. The techno-functional results support practical clinical feasibility, with substantial reductions in donning and doffing times and documented levels of therapist assistance across sessions. From a physiological standpoint, VO₂, walking speed and cost of transport remained stable between mid- and end-program assessments, while perceived exertion decreased, suggesting improved subjective tolerance with familiarization rather than progressive metabolic overload. The occurrence of a tibial plateau fracture despite adherence to current screening recommendations highlights the importance of strict bone health assessment, supporting systematic densitometry and cautious exclusion criteria in long-standing SCI. Overall, these findings support the feasibility of using Atalante to enable hands-free upright indoor mobility under supervised clinical conditions, and justify further ecologically anchored studies to characterize real-world requirements (supervision, fall-risk management) and its impact on activity and participation outcomes. Author contributions EP: Conceptualization, Methodology, Data collection, Draft writing, Review and Editing.NR: Methodology, Statistical analysis, Draft writing, Review and Editing. CB: Clinical supervision, Manuscript review. DP: Manuscript review and guidance.All authors read and approved the final manuscript. Funding No specific funding was received from public, commercial, or not-for-profit organizations for the conduct of this study. Data availability All data generated or analysed during this study are included in this published article and its supplementary information files. Additional details are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This study was approved by the French National Agency for the Safety of Medicines and Health Products (ANSM) and by the French Ethics Committee for the Protection of Individuals (Comité de Protection des Personnes, CPP), under the RCB number: 2016-A01221-50. Informed consent was obtained from all participants prior to inclusion. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement All data generated or analysed during this study are included in this published article and its supplementary information files. Additional details are available from the corresponding author on reasonable request.