2. Classification of CP

Understanding the prognosis and selecting appropriate intervention strategies for pediatric patients with CP require clinicians to use several classifications. These classifications rely on motor types, topography, the Gross Motor Function Classification System (GMFCS), and gait patterns (McIntyre et al., Reference McIntyre, Morgan, Walker and Novak2011; Peterson and Walton, Reference Peterson and Walton2016). Additionally, the Manual Ability Classification System (MACS), originally developed by Eliasson et al. (Reference Eliasson, Krumlinde-Sundholm, Rösblad, Beckung, Arner, Ohrvall and Rosenbaum2006), has been widely applied in clinical research and summarized by Paulson and Vargus-Adams (Reference Paulson and Vargus-Adams2017) to address upper extremity impairments in individuals with CP, complementing the GMFCS and other classification systems as mentioned in Paulson and Vargus-Adams (Reference Paulson and Vargus-Adams2017).

3. Topographic classification

The topographic distribution provides a common framework for classifying CP, which is based on the distribution of motor impairment. This classification helps clinicians understand which regions of the body are affected by CP (Mandaleson et al., Reference Mandaleson, Lee, Kerr and Graham2015). It can be broadly characterized as either unilateral, affecting one side of the body, or bilateral, affecting both sides (Te Velde et al., Reference Te Velde, Morgan, Novak, Tantsis and Badawi2019). Figure 1 illustrates the topographic distribution of CP (Swaroop, Reference Swaroop, Sarwark and Carl2023).

Figure 1.

Topographic distribution of CP (Swaroop, Reference Swaroop, Sarwark and Carl2023).

Table 1.

Search strategy

4. Gross Motor Function Classification System

While topographical classification describes the distribution of motor impairments in CP, it does not explain the severity of functional limitations or mobility levels. Clinicians widely use the GMFCS to evaluate gross motor function (Leviton, Reference Leviton2020; Piscitelli et al., Reference Piscitelli, Ferrarello, Ugolini, Verola and Pellicciari2021). This five-level scale categorizes individuals based on their motor function, ranging from level I, which denotes the highest degree of motor function characterized by unrestricted walking, to level V, which represents the most severe limitations in motor functions, where individuals require comprehensive assistance for mobility (Compagnone et al., Reference Compagnone, Maniglio, Camposeo, Vespino, Losito, de Rinaldis, Gennaro and Trabacca2014). For further details on the GMFCS levels, readers are referred to Compagnone et al. (Reference Compagnone, Maniglio, Camposeo, Vespino, Losito, de Rinaldis, Gennaro and Trabacca2014).

The GMFCS enables practitioners to classify functional abilities consistently over time, from initial diagnosis through subsequent assessments (Huroy et al., Reference Huroy, Behlim, Andersen, Buckley, Fehlings, Kirton, Pigeon, Mishaal, Wood, Shevell and Oskoui2022). Initially, this classification system was only used for the age group 2–12 years (Piscitelli et al., Reference Piscitelli, Ferrarello, Ugolini, Verola and Pellicciari2021). Together, the GMFCS and topographical classification systems provide a comprehensive view: topography identifies the affected body regions, while the GMFCS measures functional mobility. By integrating both systems, clinicians can predict long-term outcomes, personalize rehabilitation strategies, and evaluate treatment effectiveness (Compagnone et al., Reference Compagnone, Maniglio, Camposeo, Vespino, Losito, de Rinaldis, Gennaro and Trabacca2014).

5. Lower-limb impairments among CP subjects

Spasticity remains a prevalent concern in individuals with CP, often leading to lower limb dysfunctions (Qin et al., Reference Qin, Jiao, Zang, Pan, Shi, Guo, Zhang, Qin, Zang, Jiao and Pan2020). Clinicians frequently observe notable impairment in the distal muscles of the lower limbs across various CP types, which compromises neuromuscular control and reduces participation in daily activities (O’Brien et al., Reference O’Brien, Lichtwark, Carroll and Barber2020). In spastic diplegia, biomechanical abnormalities are prevalent, affecting almost all cases (98.4%) and spanning multiple regions from the pelvis to the ankle joint. These abnormalities include internal foot progression angle and internal and external rotation of the pelvis, hip, and ankle joints. Approximately 77% of children with spastic CP exhibit anomalies at multiple levels, and 48% of children exhibit anomalies at only one level (Simon et al., Reference Simon, Ilharreborde, Megrot, Mallet, Azarpira, Mazda, Presedo and Penneçot2015; Zhou et al., Reference Zhou, Butler and Rose2017). These anomalies manifest in various forms, such as toe-walking, ankle equinus deformity, stiff knee, and scissoring gait. Torsional abnormalities, such as femoral neck anteversion and tibial torsion, are also prevalent in infantile CP and significantly impact walking patterns, increasing the risk of falls, pain, overloading, and substantial fatigue (Frizziero et al., Reference Frizziero2022).

These biomechanical issues in the transverse plane result from a combination of static and dynamic factors, including spasticity, contractures, muscle imbalances, and excessive femoral neck anteversion. Excessive anteversion diminishes the effectiveness of hip abductors by reducing the muscular lever arms. Additionally, individuals with CP exhibit slower center-of-mass velocity at toe-off, use a wider base of support with increased step width, and have a shorter step length (Malone et al., Reference Malone, Kiernan, French, Saunders and O’Brien2016).

6. Methodology

6.1. Protocol registration and search strategy

We registered this systematic review in the PROSPERO database (Registration ID: CRD42024603481) before starting data collection and analysis. This registration ensures methodological transparency and reduces potential biases in study design and reporting. We conducted the literature search on Web of Science, PubMed, IEEE Xplore, and Scopus using the keywords “exoskeletons,” “cerebral palsy,” “kinematic,” and “robotic rehabilitation.” This study particularly focused on “robot assistive gait training” and “robotic rehabilitation”. The study selection process is illustrated in Figure 2.

Figure 2.

PRISMA flowchart for paper selection.

9. Study limitations and contradictory findings

Although lower-limb exoskeletons exhibit significant potential for improving gait and motor function in children with CP, multiple studies present contradictory results and limitations that warrant attention. As reported in the study (Yamada et al., Reference Yamada, Kadone, Shimizu and Suzuki2018), the stability of the knee joint was only improved in the supported limb, indicating a limitation in attaining bilateral improvements. The tethered knee exoskeleton (Lerner et al., Reference Lerner, Damiano and Bulea2016) exhibited increased hamstring activity in some cases, which occasionally reduced kinematic improvements, reflecting heterogeneity in muscular responses among individuals. Several studies, including the assessments of the Exoskeleton Brake Unit (Yamada et al., Reference Yamada, Kadone, Shimizu and Suzuki2018) and preliminary PediAnklebot (Michmizos et al., Reference Michmizos, Rossi, Castelli, Cappa and Krebs2015; Germanotta et al., Reference Germanotta2017) consist of small sample sizes (e.g., individual patients or small cohorts), limiting their generalizability to diverse CP populations.

The lack of long-term follow-up in the majority of studies limits their understanding of lasting treatment effects. Differences in outcome measurements (such as gait velocity compared to muscle activation) and efficacy among CP classifications (including spastic versus dyskinetic) or GMFCS levels complicate comparisons. These differences highlight the critical need for standardized methodologies, comprehensive multicenter studies, and longitudinal investigations to assess the efficacy of exoskeletons and to tackle the variability in outcomes.

10. Cost and accessibility challenges in LMICs

The cost and accessibility of robotic exoskeletons are critical issues, especially for the global CP population. As mentioned previously, considerable annual cost disparities for CP care range from $500 to $7500 in LMICs to $2,600 to $69,000 in HICs (Fang and Lerner, Reference Fang and Lerner2024). These figures reflect substantial differences not only in healthcare expenditure but also in the accessibility of advanced rehabilitation technologies.

The high production and maintenance costs of commercial exoskeletons, as well as the need for specialized setup and training facilities such as Lokomat® (Wallard et al., Reference Wallard, Dietrich, Kerlirzin and Bredin2017, Reference Wallard, Dietrich, Kerlirzin and Bredin2018; Weinberger et al., Reference Weinberger, Warken, König, Vill, Gerstl, Borggraefe, Heinen, von Kries and Schroeder2019; van Kammen et al., Reference van Kammen, Reinders-Messelink, Elsinghorst, Wesselink, Meeuwisse-de Vries, van der Woude, Boonstra and den Otter2020), HAL (Mataki et al., Reference Mataki2018; Ueno et al., Reference Ueno2019; Kuroda et al., Reference Kuroda2020, Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022, Reference Kuroda2023; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022, Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023), pose significant barriers to their widespread adoption in resource-limited settings. While more affordable and passive devices, such as the passive knee exoskeleton (Kennard et al., Reference Kennard, Kadone, Shimizu and Suzuki2022), passive pediatric leg exoskeleton (Zistatsis et al., Reference Zistatsis, Peters, Ballesteros, Feldner, Bjornson and Steele2021) offers promise in terms of simplicity and cost reduction; however, their clinical validation is currently limited.

Scaling up exoskeleton deployment in LMICs remains unrealistic with current pricing, infrastructure, and support requirements. This highlights the need for research and policy focused not only on technological advancement but also on cost-effective, accessible, and locally manufacturable solutions. In addition, further evaluation of device durability, availability of technical support, and reimbursement frameworks is an essential step toward equitable access.

11. Artificial Intelligence integration and real-world deployment

Devices like the Biomotum Spark (Tagoe et al., Reference Tagoe, Fang, Williams and Lerner2023) and the portable pediatric knee exoskeleton (Zhang et al., Reference Zhang, Zhu, Huang, Yu, Huang, Lopez-Sanchez, Devine, Abdelhady, Zheng, Bulea and Su2024) show how the incorporation of artificial intelligence (AI) greatly enhance the functionality of exoskeletons. Based on gait patterns and neuromuscular signals, such as electromyography (EMG) and electroencephalography (EEG), these devices use machine learning algorithms to dynamically modify torque assistance in real-time (Tagoe et al., Reference Tagoe, Fang, Williams and Lerner2023; Zhang et al., Reference Zhang, Zhu, Huang, Yu, Huang, Lopez-Sanchez, Devine, Abdelhady, Zheng, Bulea and Su2024). However, there are several challenges in integrating AI for real-world use. The heterogeneity of cerebral palsy limits personalization due to small datasets, which are common in pediatric research, leading to overfitting and reducing generalizability across age groups or different levels of the Gross Motor Function Classification System.

The ethical considerations related to AI interventions encompass algorithmic biases, constraints on resources, and obstacles in communication, which present challenges to rehabilitation outcomes and exacerbate disparities in environments with limited resources (Bulea et al., Reference Bulea, Chen and Damiano2020). Data management concerns involve General Data Protection Regulation (GDPR) compliance and protecting pediatric patients’ privacy. Implementing stringent AI validation, ethical standards, and affordable devices is crucial for safe, efficient, and fair deployment (Tibebu, Reference Tibebun.d.; Balgude et al., Reference Balgude, Gite, Pradhan and Lee2024; Chng et al., Reference Chng, Tern, Lee, Cheng, Kapur, Eriksson, Chong and Savulescu2025).

12. Advancements in training interfaces

Notable advancements have been achieved by transitioning from passive mechanical support to active and sensor-driven integration of training interfaces. Gamification is emerging as a prominent feature aimed at improving motor learning. Exoskeleton incorporates interactive elements, including virtual reality (VR) (Wallard et al., Reference Wallard, Dietrich, Kerlirzin and Bredin2017, Reference Wallard, Dietrich, Kerlirzin and Bredin2018, Weinberger et al., Reference Weinberger, Warken, König, Vill, Gerstl, Borggraefe, Heinen, von Kries and Schroeder2019, van Kammen et al., Reference van Kammen, Reinders-Messelink, Elsinghorst, Wesselink, Meeuwisse-de Vries, van der Woude, Boonstra and den Otter2020) and game-based rehabilitation (Bulea et al., Reference Bulea, Lerner, Gravunder and Damiano2017). In contrast, earlier exoskeleton devices like Lokomat® (Wallard et al., Reference Wallard, Dietrich, Kerlirzin and Bredin2017, Reference Wallard, Dietrich, Kerlirzin and Bredin2018; Digiacomo et al., Reference Digiacomo, Tamburin, Tebaldi, Pezzani, Tagliafierro, Casale and Bartolo2019; Weinberger et al., Reference Weinberger, Warken, König, Vill, Gerstl, Borggraefe, Heinen, von Kries and Schroeder2019; van Kammen et al., Reference van Kammen, Reinders-Messelink, Elsinghorst, Wesselink, Meeuwisse-de Vries, van der Woude, Boonstra and den Otter2020) and MIT-Skywalker (Susko et al., Reference Susko, Swaminathan and Krebs2016) relied on more rudimentary biofeedback mechanisms. The incorporation of immersive environments for practice has significantly improved patient engagement and enthusiasm, as demonstrated by the utilization of devices such as WAKE-Up Exoskeleton (Patané et al., Reference Patané, Rossi, Sette, Taborri and Cappa2017). The implementation of gamified audiovisual feedback in conjunction with powered knee exoskeletons such as PediAnklebot (Michmizos et al., Reference Michmizos, Rossi, Castelli, Cappa and Krebs2015; Germanotta et al., Reference Germanotta2017), ProGait (McDaid, Reference McDaid2017) Tethered Knee Exoskeleton (Lerner et al., Reference Lerner, Damiano and Bulea2016), and Powered Knee Exoskeleton (Lerner et al., Reference Lerner, Damiano, Park, Gravunder and Bulea2016, Reference Lerner, Damiano and Bulea2017a,Reference Lerner, Damiano and Buleab; Bulea et al., Reference Bulea, Lerner, Gravunder and Damiano2017) has demonstrated advantages in enhancing cortical activation and fostering voluntary motor engagement. Furthermore, the interactive interface, which constitutes biofeedback mechanisms through EMG, EEG, Audiovisual feedback, and real-time torque estimation, allows the implementation of personalized rehabilitation strategies (Lerner et al., Reference Lerner2018, Reference Lerner, Conner and Remec2019a,Reference Lerner, Harvey and Lawsonb; Mataki et al., Reference Mataki2018; Gasparri et al., Reference Gasparri, Luque and Lerner2019; Ueno et al., Reference Ueno2019; Conner et al., Reference Conner, Luque and Lerner2020, Reference Conner, Schwartz and Lerner2021, Reference Conner, Spomer, Steele and Lerner2023; Kuroda et al., Reference Kuroda2020; Orekhov et al., Reference Orekhov, Fang, Luque and Lerner2020; Fang et al., Reference Fang, Orekhov and Lerner2021; Fang and Lerner, Reference Fang and Lerner2021; Harvey et al., Reference Harvey, Conner and Lerner2021; Orekhov et al., Reference Orekhov, Fang, Cuddeback and Lerner2021; Conner and Lerner, Reference Conner and Lerner2022; Fang et al., Reference Fang, Orekhov and Lerner2022; Fang and Lerner, Reference Fang and Lerner2022; Kuroda et al., Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022; Harshe et al., Reference Harshe, Williams, Hocking and Lerner2023; Kuroda et al., Reference Kuroda2023; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023; Fang and Lerner, Reference Fang and Lerner2024). HAL (Mataki et al., Reference Mataki2018; Ueno et al., Reference Ueno2019; Kuroda et al., Reference Kuroda2020, Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022, Reference Kuroda2023; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022, Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023) and Lokomat (Wallard et al., Reference Wallard, Dietrich, Kerlirzin and Bredin2017, Reference Wallard, Dietrich, Kerlirzin and Bredin2018; Weinberger et al., Reference Weinberger, Warken, König, Vill, Gerstl, Borggraefe, Heinen, von Kries and Schroeder2019; van Kammen et al., Reference van Kammen, Reinders-Messelink, Elsinghorst, Wesselink, Meeuwisse-de Vries, van der Woude, Boonstra and den Otter2020) incorporate VR and CVC to refine proprioceptive training.

13. Changes in clinical studies

The design of clinical studies involving exoskeletons has progressed from initial pilot testing to larger-scale clinical trials and feasibility assessments. Initial studies, such as the exoskeleton brake unit (Yamada et al., Reference Yamada, Kadone, Shimizu and Suzuki2018), assessed the stability of the knee on a single-patient basis, with limited outcome measures. The preliminary investigations were limited in scale and focused on evaluating safety and basic functionality, for example, the initial pilot studies for the PediAnklebot (Michmizos et al., Reference Michmizos, Rossi, Castelli, Cappa and Krebs2015; Germanotta et al., Reference Germanotta2017), powered knee exoskeleton (Lerner et al., Reference Lerner, Damiano, Park, Gravunder and Bulea2016, Reference Lerner, Damiano and Bulea2017a,Reference Lerner, Damiano and Buleab; Bulea et al., Reference Bulea, Lerner, Gravunder and Damiano2017), and untethered ankle exoskeleton (Gasparri et al., Reference Gasparri, Bair, Libby and Lerner2018; Orekhov et al., Reference Orekhov, Fang, Cuddeback and Lerner2021; Conner and Lerner, Reference Conner and Lerner2022; Fang and Lerner, Reference Fang and Lerner2022; Conner et al., Reference Conner, Spomer, Steele and Lerner2023; Harshe et al., Reference Harshe, Williams, Hocking and Lerner2023; Fang and Lerner, Reference Fang and Lerner2024). Feasibility assessments were performed for CP Walker (Bayón et al., Reference Bayón2016a,Reference Bayónb), WAKE-Up Exoskeleton (Patané et al., Reference Patané, Rossi, Sette, Taborri and Cappa2017), and HAL (Mataki et al., Reference Mataki2018; Ueno et al., Reference Ueno2019; Kuroda et al., Reference Kuroda2020, Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022, Reference Kuroda2023; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022, Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023), whereas extensive clinical trials were carried out for Lokomat® Pediatric (Wallard et al., Reference Wallard, Dietrich, Kerlirzin and Bredin2017, Reference Wallard, Dietrich, Kerlirzin and Bredin2018; Weinberger et al., Reference Weinberger, Warken, König, Vill, Gerstl, Borggraefe, Heinen, von Kries and Schroeder2019; van Kammen et al., Reference van Kammen, Reinders-Messelink, Elsinghorst, Wesselink, Meeuwisse-de Vries, van der Woude, Boonstra and den Otter2020), Angel Legs (Kim et al., Reference Kim2021), and HWA (Kawasaki et al., Reference Kawasaki, Ohata, Yoshida, Yokoyama and Yamada2020). Extensive RCTs have recently been undertaken, incorporating control groups and longitudinal measurements for Lokomat® Pediatric (Wallard et al., Reference Wallard, Dietrich, Kerlirzin and Bredin2017, Reference Wallard, Dietrich, Kerlirzin and Bredin2018; Digiacomo et al., Reference Digiacomo, Tamburin, Tebaldi, Pezzani, Tagliafierro, Casale and Bartolo2019; Weinberger et al., Reference Weinberger, Warken, König, Vill, Gerstl, Borggraefe, Heinen, von Kries and Schroeder2019; van Kammen et al., Reference van Kammen, Reinders-Messelink, Elsinghorst, Wesselink, Meeuwisse-de Vries, van der Woude, Boonstra and den Otter2020), CP Walker (Bayón et al., Reference Bayón2016a,Reference Bayónb), HAL (Mataki et al., Reference Mataki2018; Ueno et al., Reference Ueno2019; Kuroda et al., Reference Kuroda2020, Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022, Reference Kuroda2023; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022, Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023), Angel Legs (Kim et al., Reference Kim2021), and HWA (Kawasaki et al., Reference Kawasaki, Ohata, Yoshida, Yokoyama and Yamada2020). Larger-scale multicenter studies are essential to improve the generalizability of the findings. The inclusion of standardized assessment outcome measures, including the GMFCS, Physiological Cost Index, and Six-Minute Walk Test (6MWT), has enabled a more objective evaluation of effectiveness (Bayón et al., Reference Bayón2016a,Reference Bayónb; Lerner et al., Reference Lerner, Damiano and Bulea2016; Susko et al., Reference Susko, Swaminathan and Krebs2016; Bulea et al., Reference Bulea, Lerner, Gravunder and Damiano2017; McDaid, Reference McDaid2017; Wallard et al., Reference Wallard, Dietrich, Kerlirzin and Bredin2017, Reference Wallard, Dietrich, Kerlirzin and Bredin2018; Lerner et al., Reference Lerner2018, Reference Lerner, Conner and Remec2019a,Reference Lerner, Harvey and Lawsonb; Mataki et al., Reference Mataki2018; Gasparri et al., Reference Gasparri, Luque and Lerner2019; Ueno et al., Reference Ueno2019; Weinberger et al., Reference Weinberger, Warken, König, Vill, Gerstl, Borggraefe, Heinen, von Kries and Schroeder2019; Bulea et al., Reference Bulea, Chen and Damiano2020, Reference Bulea, Molazadeh, Thurston and Damiano2022; Conner et al., Reference Conner, Luque and Lerner2020a,Reference Conner, Remec, Orum, Frank and Lernerb, Reference Conner, Schwartz and Lerner2021, Reference Conner, Spomer, Steele and Lerner2023; Kuroda et al., Reference Kuroda2020, Reference Kuroda2023; Orekhov et al., Reference Orekhov, Fang, Luque and Lerner2020; van Kammen et al., Reference van Kammen, Reinders-Messelink, Elsinghorst, Wesselink, Meeuwisse-de Vries, van der Woude, Boonstra and den Otter2020; Chen et al., Reference Chen2021; Delgado et al., Reference Delgado2021; Fang et al., Reference Fang, Orekhov and Lerner2021; Fang and Lerner, Reference Fang and Lerner2021, Reference Fang and Lerner2024; Harvey et al., Reference Harvey, Conner and Lerner2021; Kim et al., Reference Kim2021; Orekhov et al., Reference Orekhov, Fang, Cuddeback and Lerner2021; Conner and Lerner, Reference Conner and Lerner2022; Fang et al., Reference Fang, Orekhov and Lerner2022; Fang and Lerner, Reference Fang and Lerner2022; Kuroda et al., Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022, Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023; Harshe et al., Reference Harshe, Williams, Hocking and Lerner2023; Tagoe et al., Reference Tagoe, Fang, Williams and Lerner2023; Zhang et al., Reference Zhang, Zhu, Huang, Yu, Huang, Lopez-Sanchez, Devine, Abdelhady, Zheng, Bulea and Su2024).

14. Improvement in outcomes over time

The advancement of exoskeleton technology has correspondingly improved clinical outcomes. Initial investigations concentrated on feasibility, revealing only limited enhancements in mobility and muscle activation. Nonetheless, advancements in more sophisticated control algorithms, lightweight materials, and adaptive training environments have led to notable enhancements in rehabilitation outcomes.

15. Trends and existing gaps

Several trends were observed in the context of research on the lower-limb exoskeleton for CP. New research in these devices aims at portability and reduced weight, as seen in the untethered exoskeleton, including Biomotum Spark (Tagoe et al., Reference Tagoe, Fang, Williams and Lerner2023), HWA (Kawasaki et al., Reference Kawasaki, Ohata, Yoshida, Yokoyama and Yamada2020), and untethered ankle exoskeletons (Gasparri et al., Reference Gasparri, Bair, Libby and Lerner2018; Orekhov et al., Reference Orekhov, Fang, Cuddeback and Lerner2021; Conner and Lerner, Reference Conner and Lerner2022; Fang and Lerner, Reference Fang and Lerner2022, Reference Fang and Lerner2024; Conner et al., Reference Conner, Spomer, Steele and Lerner2023; Harshe et al., Reference Harshe, Williams, Hocking and Lerner2023) for real-world settings. The implementation of AI-driven adaptive control systems is being integrated into many devices, including the CP Walker (Bayón et al., Reference Bayón2016a,Reference Bayónb), WAKE-Up Exoskeleton (Patané et al., Reference Patané, Rossi, Sette, Taborri and Cappa2017), and HAL (Mataki et al., Reference Mataki2018; Ueno et al., Reference Ueno2019; Kuroda et al., Reference Kuroda2020, Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022, Reference Kuroda2023; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022, Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023), permitting real-time gait modifications to enhance the experience of patients and improve motor learning. Similarly, another trend is to incorporate multisensory integration, advanced haptic, and neurofeedback to improve the effectiveness of training and engagement of patients during interventions. Devices such as the WAKE-Up Exoskeleton (Patané et al., Reference Patané, Rossi, Sette, Taborri and Cappa2017) have successfully provided leverage toward these interactive features to improve rehabilitation outcomes.

Despite the evidence of short-term benefits, there is still a significant lack of knowledge regarding the long-term efficacy of exoskeleton application in CP rehabilitation. To address the diversity of neuromotor impairments among populations impacted by CP, device customization is essential, as demonstrated in HAL (Mataki et al., Reference Mataki2018; Ueno et al., Reference Ueno2019; Kuroda et al., Reference Kuroda2020, Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022, Reference Kuroda2023; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022, Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023) and CP Walker (Bayón et al., Reference Bayón2016a,Reference Bayónb).

16. Future directions

The new technologies in exoskeletons should be focused on several major developments. Of particular significance is cable-driven actuation, which can be used to minimize the overall weight of exoskeletons, while still interacting with the human subjects, for example, cable-driven active leg exoskeleton (C-ALEX) (Hidayah et al., Reference Hidayah, Bishop, Jin, Chamarthy, Stein and Agrawal2020), tethered pelvic assist device (TPAD) (Kang et al., Reference Kang, Martelli, Vashista, Martinez-Hernandez, Kim and Agrawal2017), mobile tethered pelvic assist device (mTPAD) (Martelli et al., Reference Martelli, Luo, Kang, Kang, Fahn and Agrawal2017), and robotic upright stand trainer (RobUST) (Rejc et al., Reference Rejc, Zaccaron, Bowersock, Pisolkar, Ugiliweneza, Forrest, Agrawal, Harkema and Angeli2024). Technologies such as the WAKE-Up Exoskeleton (Patané et al., Reference Patané, Rossi, Sette, Taborri and Cappa2017) and Angel Legs (Kim et al., Reference Kim2021) are in the process of integrating such mechanisms. Of particular interest is the implementation of AI-based assistance, wherein personalized machine learning methods for gait optimization can make such devices more flexible for each individual. Moreover, cost-effective production techniques are being studied to make exoskeletons available to rehabilitation centers, thereby making them more marketable.

There is also greater interest in making the user more comfortable with more advanced control schemes so that exoskeletons can provide more intuitive assistance based on real-time feedback from the user. This requires hybrid actuation schemes that combine active and passive elements to make the exoskeleton more energy-efficient and less exhausting for the user. There is also likely to be advanced material science that makes exoskeletons stronger, more flexible, and evenly weight-bearing so that they are more comfortable and usable for longer periods.

Eventually, future technologies should be focused on expanding devices to reach more people, including powered knee exoskeletons (Lerner et al., Reference Lerner, Damiano, Park, Gravunder and Bulea2016, Reference Lerner, Damiano and Bulea2017a,Reference Lerner, Damiano and Buleab; Bulea et al., Reference Bulea, Lerner, Gravunder and Damiano2017), untethered exoskeletons (Gasparri et al., Reference Gasparri, Bair, Libby and Lerner2018;Orekhov et al., Reference Orekhov, Fang, Cuddeback and Lerner2021; Conner and Lerner, Reference Conner and Lerner2022; Fang and Lerner, Reference Fang and Lerner2022, Reference Fang and Lerner2024; Conner et al., Reference Conner, Spomer, Steele and Lerner2023; Harshe et al., Reference Harshe, Williams, Hocking and Lerner2023) at the ankles, passive leg exoskeletons for CP, and whole-body rehabilitation devices like the HAL (Mataki et al., Reference Mataki2018; Ueno et al., Reference Ueno2019; Kuroda et al., Reference Kuroda2020, Reference Kuroda, Mutsuzaki, Nakagawa, Yoshikawa, Takahashi, Mataki, Takeuchi, Iwasaki and Yamazaki2022, Reference Kuroda2023; Moll et al., Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2022, Reference Moll, Kessel, Bonetto, Stresow, Herten, Dudda and Adermann2023). They will be more personalized, effective, and available to more people with neuromotor impairment.

Importantly, future efforts must also address the affordability and accessibility of these technologies in LMICs, where the prevalence of CP is often higher and resources are limited (Kakooza-Mwesige et al., Reference Kakooza-Mwesige, Andrews, Peterson, Mangen, Eliasson and Forssberg2017).

Overall, exoskeleton technology has advanced greatly, but further research needs to be conducted to make these usable, effective in real-world use, and accepted in clinical and home settings.