Effectiveness and usability of artificial intelligence-powered assistive technologies in Supporting daily activities of children with cerebral palsy: a systematic review

Abstract

Background

Cerebral Palsy (CP) is the main cause of motor disabilities in childhood, necessitating innovative approaches to rehabilitation and assistive technology (AT). Simultaneously, artificial intelligence (AI) is increasingly being integrated into devices to create more adaptive, personalized, and effective AT. This systematic review aimed to evaluate the effectiveness and usability of AI-powered assistive technologies designed to support daily activities and rehabilitation in children with CP.

Materials and methods

Five databases, including Scopus, Web of Science, PubMed, Embase, and IEEE Xplore, were systematically searched, and 23 articles were included in the final analysis. Articles were identified, selected, and categorized into emerging thematic areas based on the primary function and application of the technology.

Results

Five key thematic topics were identified: 1) AI-driven motor rehabilitation and gait training for functional mobility; 2) intelligent assessment and monitoring systems for clinical decision support; 3) AI-supported communication, social interaction, and intention recognition tools; 4) gamified and virtual reality–based interventions to enhance engagement and usability; and 5) smart assistive systems supporting daily living and independent mobility. The findings demonstrate a strong trend toward the application of AI technologies in personalized, engaging, and data-driven interventions for children with CP. However, the field is predominantly in the proof-of-concept stage, with limitations including small sample sizes, lack of long-term clinical validation, challenges in user-centered design, and usability for children with CP.

Conclusion

AI-powered assistive technologies hold significant potential for transforming the care of children with CP by enabling highly personalized and engaging interventions. To actualize this potential, future work must realize that practical application remains challenging owing to limited clinical validation, technological integration, and usability barriers for children with CP. Future research must prioritize user-centered design and multidisciplinary collaboration to ensure that AI and robotic advancements improve the usability and quality of life for children with CP.

KEY MESSAGES

1. This is a systematic review of 23 research articles evaluating the effectiveness and usability of artificial intelligence-powered assistive technologies for supporting daily activities in children with cerebral palsy.

2. AI is enabling a new generation of assistive technologies for children with cerebral palsy that are adaptive, personalized, and focused on improving function across daily life activities, from motor rehabilitation and communication to independent mobility.

3. The successful adoption of AI technologies depend on a user-centered, multidisciplinary approach that closely involves children, families, and clinicians to ensure solutions are not only technologically advanced but also practical, engaging, and clinically meaningful.

1. Introduction

Nearly 240 million children experience some degree of mobility disability, whether present at birth or acquired later in life [1]. Moreover, the number of children born with mobility disabilities has been increasing [2]. Mobility disabilities in children may result from various causes, including congenital conditions, traumatic injuries, and chronic diseases. Cerebral palsy (CP) is the leading cause of mobility disability in children [3]. Globally, approximately four out of every 1,000 children are born with CP [4]. CP is a neurological disorder caused by abnormal brain development and is often accompanied by secondary impairments such as cognitive, language, and visual difficulties, which can significantly limit children’s daily activities. It is typically diagnosed during infancy or early childhood due to signs such as delayed motor development, abnormal muscle tone, or atypical reflexes [5].

CP is divided into several subtypes in children, with spastic CP being the most prevalent. This subtype is characterized by increased muscle tone in the limbs, resulting in spasticity and the characteristic scissoring posture of the lower extremities. The second most common subtype is extrapyramidal or athetoid CP, which typically presents as hypotonia in early infancy and later progresses to choreoathetoid movements and dystonia as the child grows. Ataxic CP is marked by hypotonia and impaired coordination, while mixed CP presents overlapping features of multiple clinical subtypes [6]. Children diagnosed with CP often exhibit a broad spectrum of symptoms that affect daily functioning and limit participation in age-appropriate activities, making even simple tasks challenging. In addition, many children experience comorbidities such as speech impairments, seizures, and cognitive difficulties, which further complicate care and rehabilitation [7]. The variability in symptom severity and the diverse needs of each child present significant challenges for caregivers and healthcare professionals in providing effective, individualized support [8]. Consequently, CP not only impacts children’s physical abilities but also affects their long-term independence and quality of life, highlighting the urgent need for innovative and inclusive care solutions.

Childcare for children with disabilities, particularly those with CP, has long been a major focus for designers, healthcare professionals, educators, sanitation facilities, and protection services worldwide. These efforts aim to ensure that children with disabilities receive adequate support and to address the mental health needs of their caregivers [9]. Daily activities that promote early childhood development are essential for helping children achieve their full physical, cognitive, and social potential, ultimately enabling them to thrive and contribute to a sustainable future [10]. Despite increasing global awareness, many children with disabilities – especially those diagnosed with CP – continue to receive insufficient care and have limited opportunities for functional development, particularly children from low-income households [11]. This reflects a pressing global challenge that demands coordinated attention and intervention.

To address the challenges faced by children with movement disabilities caused by CP, assistive technology (AT) interventions are frequently recommended. In this review, AT refers to devices that support functional independence in daily activities, while rehabilitation technologies refer to therapeutic systems designed to improve motor or cognitive functions through training or intervention. AI-powered systems may enhance both categories by enabling adaptive feedback, intelligent monitoring, and personalized support [12,13]. AT encompasses a wide range of devices and related services designed to support or enhance an individual’s abilities in areas such as cognition, communication, hearing, mobility, self-care, and vision, thereby promoting overall health and active participation in daily life [14]. Existing research on AT has made valuable contributions to the rehabilitation and care of children with CP, including the development of mobility aids, communication tools, and therapeutic devices [15]. Although AT provides meaningful opportunities for children with CP to improve their mobility and independence – offering benefits such as enhanced movement, increased participation, and greater autonomy in daily tasks – current design solutions still exhibit notable limitations. Many available devices are overly complex, insufficiently adaptable, or poorly aligned with the specific developmental and functional needs of children, leaving many unable to fully benefit from these technologies. The effectiveness and usability of AT for children with CP therefore require significant improvement [16]. In addition, usability challenges, high costs, and limited integration into daily routines often result in low adoption rates among families and caregivers [17]. This persistent gap highlights the inadequacy of current solutions and underscores the urgent need for more personalized, accessible, and effective approaches to AT that can truly empower children with CP to achieve greater independence in their daily activities.

Current applications of AT for children with CP have not fully addressed their dynamic and individualized needs, and usability remains a significant area requiring improvement [18]. At the same time, the rapid emergence of advanced technologies – such as artificial intelligence (AI) and the Internet of Things (IoT) – has gained increasing attention. IoT refers to a network of connected devices that collect and share data via the internet to enable smart monitoring and control [19]. AI involves the simulation of human intelligence in machines capable of learning, reasoning, and making decisions [20]. Together, these interconnected technologies enable real-time data collection, intelligent analysis, and immersive interactions that bridge digital information and the physical world [21]. The integration of IoT and AI represents a new generation of smart assistive technologies capable of enhancing usability and facilitating self-management among children with CP. These advances provide increased opportunities to develop smart, adaptive, and user-centered assistive solutions tailored to children’s unique needs [22]. Overall, such technologies hold strong potential for delivering personalized rehabilitation support, real-time monitoring, and seamless integration into the daily routines and environments of children with CP [23]. Despite extensive research on AT for children with CP, there remains a notable gap regarding the use of emerging technologies – specifically IoT and AI – within assistive solutions, commonly referred to as smart assistive technology (SAT). Research focusing on SAT to improve usability and personalized services for children with CP is still limited. Given the growing demand and expanding market, investigating innovative SAT designs for children with CP is increasingly necessary.

Although previous studies have explored assistive technologies and rehabilitation devices for children with cerebral palsy, most existing reviews focus primarily on traditional rehabilitation tools or general assistive devices. Few reviews have systematically examined the integration of emerging technologies such as artificial intelligence (AI) and the Internet of Things (IoT) within assistive systems, particularly in the context of pediatric rehabilitation. In addition, prior reviews rarely evaluate usability outcomes and real-world applicability of these technologies for children with CP. As a result, the current state of AI-enabled smart assistive technologies for pediatric populations remains insufficiently synthesized, highlighting the need for a comprehensive review focusing on the effectiveness and usability of these emerging systems.

Despite the growing number of studies on AI- and IoT-integrated AT, their functional effectiveness and real-world usability for children with CP remain poorly understood. Therefore, this systematic review aims to evaluate the effectiveness and usability of these technologies in improving the daily functional activities of children with CP in the era of smart technology and artificial intelligence. By synthesizing the current evidence, this review provides valuable insights to inform future research and guide the design of assistive technologies that enhance quality of life and enable children with CP to benefit more fully from technological advancements.

2. Materials and methods

The protocol for this review was registered in the PROSPERO database (registered ID: CRD420251246555) and conducted in strict accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [24,25]. The completed PRISMA checklist is provided in Supplementary Material 1.

2.1. Search strategy

This study employed a systematic review approach to analyze relevant articles that focused on a specific topic during a limited time period: SAT for children with CP in the AI-driven AT context. The initial systematic review process was conducted using the following databases: Scopus, Web of Science (WoS), PubMed, Embase, and Institute of Electrical and Electronics Engineers Xplore (IEEE). IEEE is included in this medical review because it improves coverage, as AI and IoT are grounded in computer and engineering sciences. Subsequently, relevant articles were identified using the specific keywords outlined in the search strategy. Following the PECO/PICO/PIO framework – representing person, exposure (intervention), comparison, and outcome [26], the search was conducted using defined search terms: P: (‘child*’ OR ‘young baby’ OR ‘toddler’ OR ‘paediatric’ OR ‘pediatric’) AND (‘cerebral pals*’ OR ‘little disease’ OR ‘infantile pals*’ OR ‘diplegia’ OR ‘spastic’ OR ‘spastic diplegia*’); I: ‘IoT’ OR ‘internet of thing*’ OR ‘artificial intelligence’ OR ‘smart technolog*’; O: ‘AT’ OR ‘assistive technolog*’ OR ‘assistive device*’ OR ‘AD’ OR ‘assistive product*’ OR ‘product*’ OR ‘product* design’ OR design.

The complete search strings and PICO-based search strategy used across the databases are provided in Supplementary Material 2. Keywords (‘child*’ OR ‘young baby’ OR ‘toddler’ OR ‘paediatric’ OR ‘pediatric’) AND (‘cerebral pals*’ OR ‘little disease’ OR ‘infantile pals*’ OR ‘diplegia’ OR ‘spastic’ OR ‘spastic diplegia*’) AND (‘IoT’ OR ‘internet of thing*’ OR ‘AIoT’ OR ‘artificial intelligence of thing*’ OR ‘artificial intelligence’ OR AI OR ‘smart technolog*’) AND (‘AT’ OR ‘assistive technolog*’ OR ‘assistive device*’ OR ‘AD’ OR ‘assistive product*’ OR ‘product*’ OR ‘product* design’ OR design) are keyed in to search for specific research in databases (Scopus, WoS, PubMed, Embase, and IEEE). The search period spanned from 2015 to 2025. This timeframe was selected to capture the most recent developments in the integration of AI and IoT into assistive technologies, reflecting the rapid advancement of AI-driven rehabilitation technologies over the past decade. From the search results listed in Table 1, 526 articles were identified for this review.

Table 1.

Search strings from scopus, WoS, PubMed, embase and IEEE.

SCOPUS	TITLE-ABS-KEY (“child*” OR “young baby” OR “toddler” OR “paediatric” OR “pediatric”) AND TITLE-ABS-KEY (“cerebral pals*” OR “little disease” OR “infantile pals*”OR “diplegia” OR “spastic” OR “spastic diplegia*”) AND TITLE-ABS-KEY (“IoT” OR “internet of thing*” OR “artificial intelligence” OR AI OR “AIoT” OR “artificial intelligence or thing*” OR “smart technolog*”) AND TITLE-ABS-KEY (“AT” OR “assistive technolog*” OR “assistive device*” OR “AD” OR “assistive product*” OR “product*” OR “product* design” OR design)	68 results
WoS	(TS=(“child*” OR “young baby” OR “toddler” OR “paediatric” OR “pediatric”) AND TS=(“cerebral pals*” OR “little disease” OR “infantile pals*” OR “diplegia” OR “spastic” OR “spastic diplegia*”) AND TS=(“IoT” OR “internet of thing*” OR “artificial intelligence” OR AI OR “AIoT” OR “artificial intelligence or thing*” OR “smart technolog*”) AND TS=(“AT” OR “assistive technolog*” OR “assistive device*” OR “AD” OR “assistive product*” OR “product*” OR “product* design” OR design))	141 results
PubMed	(“child*” OR “young baby” OR “toddler” OR “paediatric” OR “pediatric”) AND (“cerebral pals*” OR “little disease” OR “infantile pals*” OR “diplegia” OR “spastic” OR “spastic diplegia*”) AND (“IoT” OR “internet of thing*” OR “artificial intelligence” OR AI OR “AIoT” OR “artificial intelligence or thing*” OR “smart technolog*”) AND (“AT” OR “assistive technolog*” OR “assistive device*” OR “AD” OR “assistive product*” OR “product*” OR “product* design” OR design)	109 results
Embase	(“child*” OR “young baby” OR “toddler” OR “paediatric” OR “pediatric”) AND (“cerebral pals*” OR “little disease” OR “infantile pals*” OR “diplegia” OR “spastic” OR “spastic diplegia*”) AND (“IoT” OR “internet of thing*” OR “artificial intelligence” OR AI OR “AIoT” OR “artificial intelligence or thing*” OR “smart technolog*”) AND (“AT” OR “assistive technolog*” OR “assistive device*” OR “AD” OR “assistive product*” OR “product*” OR “product* design” OR design)	201 results
IEEE	(“child*” OR “young baby” OR “toddler” OR “paediatric” OR “pediatric”) AND (“cerebral pals*” OR “little disease” OR “infantile pals*” OR “diplegia” OR “spastic” OR “spastic diplegia*”) AND (“IoT” OR “internet of thing*” OR “artificial intelligence” OR AI OR “AIoT” OR “artificial intelligence or thing*” OR “smart technolog*”) AND (“AT” OR “assistive technolog*” OR “assistive device*” OR “AD” OR “assistive product*” OR “product*” OR “product* design” OR design)	37 results

2.2. Eligibility criteria

Studies were evaluated for eligibility according to predefined inclusion and exclusion criteria. Studies were included if they met the following criteria:

1. Studies involving children (aged 4–12) diagnosed with CP [9,12,27]

2. Studies that applied and assessed the usability or effectiveness of assistive technologies integrated with AI and IoT, with the purpose of improving mobility and daily functional activities among children with cerebral palsy

3. Full-text articles that are available in English

Studies were excluded if they met any of the following criteria:

1. Not focused on children with mobility disability or the CP population

2. Primarily investigated medical or surgical interventions for children’s disability without assistive technology

3. Studies that did not explicitly focus on the integration or application of AI and the IoT within assistive technologies designed for children with CP

4. Publications not written in English

5. Secondary sources, including review papers, opinion articles, editorials, or perspective pieces

6. Grey literature, such as theses, dissertations, unpublished reports, or conference proceedings

7. No available full text

2.3. Study selection

During the article selection process, 556 articles were initially identified and carefully filtered based on the inclusion and exclusion criteria by the first and second authors. Duplicate records were removed before the screening stage. For the remaining articles, the full texts of each study were screened and evaluated by the first and second authors to ensure the accuracy and consistency of the selection process.

2.4. Assessment of study quality

Given the heterogeneity of the included studies, methodological quality was assessed using the Mixed Methods Appraisal Tool (MMAT) 2018 version [28]. The MMAT is developed for systematic reviews involving diverse research methodologies, including qualitative, quantitative, and mixed-methods research. It allows the methodological quality of studies to be evaluated across five study design categories: qualitative studies, randomized controlled trials, non-randomized studies, quantitative descriptive studies, and mixed-methods studies.

The MMAT assessment evaluated methodological rigor, appropriateness of study design, sampling strategy, data collection procedures, outcome measurement, risk of bias, and interpretation of findings. Two reviewers independently conducted the quality assessment, and any disagreements were resolved through discussion to ensure consistency and reliability [29].

Quality appraisal outcomes were not used as exclusion criteria; rather, they informed interpretive caution during the synthesis and discussion of findings. Detailed MMAT assessment results for all included studies are provided in Supplementary Material 3.

2.5. Assessment of the overall certainty of evidence

To complement the risk-of-bias and study quality assessments, the overall certainty of evidence across the included studies was evaluated using the GRADE approach framework. This approach assesses the strength of the overall body of evidence by considering factors such as study design, risk of bias, consistency of findings, indirectness, and imprecision. Based on these criteria, the certainty of evidence is categorized as high, moderate, low, or very low [30].

However, due to the heterogeneity of study designs, intervention types, and outcome measures among the included studies, the GRADE assessment was applied cautiously and interpreted with consideration of these limitations. Therefore, the GRADE ratings should be interpreted as indicative rather than definitive estimates of evidence certainty.

3. Results

3.1. Screening the article results

The PRISMA screening flowchart is presented in Figure 1. A total of 556 articles were initially identified through a systematic search, of which 132 duplicate records were removed, leaving 424 articles for screening. The first stage involved title screening, after which 214 articles with irrelevant titles were excluded from the review. Subsequently, articles with abstracts that did not align with the research focus were removed, resulting in the exclusion of 69 papers. The remaining articles were evaluated according to the inclusion and exclusion criteria, leading to the removal of 92 articles. Finally, 23 articles met the eligibility criteria and were included in this review. The full list of the included studies is provided in Supplementary Material 4.

Figure 1.

Flow chart of search strategy based on PRISMA flow diagram.

The methodological quality assessment results for the 23 included studies are presented in Table 2 and were evaluated using the Mixed Methods Appraisal Tool (MMAT) version 2018. The assessment focused on key methodological criteria such as the appropriateness of the study design, adequacy of measurements and data collection, completeness of outcome data, integration of qualitative and quantitative components, and consideration of potential confounding factors.

Table 2.

Mixed methods Appraisal tools (MMAT) scoring ratings.

MMAT checklist evaluation criteria (‘Y’ =Yes, ‘N’ =No, and ‘U’ =Can’t tell)
	Screening questions		1. Qualitative					3. Quantitative non-randomized					4. Quantitative descriptive					5. Mixed methods					Score (%)
Authors (Year)	S1	S2	1.1	1.2	1.3	1.4	1.5	3.1	3.2	3.3	3.4	3.5	4.1	4.2	4.3	4.4	4.5	5.1	5.2	5.3	5.4	5.5
Qualitative studies
(Kent et al. 2021) [31]	Y	Y	Y	U	Y	U	Y																60%
(Sabater-Gárriz et al. 2025) [32]	Y	Y	Y	Y	U	U	Y																60%
(Hsieh et al. 2021) [33]	Y	Y	Y	Y	Y	Y	Y																100%
(Encarnação et al. 2017) [34]	Y	Y	Y	Y	Y	U	Y																80%
Quantitative non-randomized studies
(Buitrago et al. 2020) [35]	Y	Y						N	Y	Y	N	Y											60%
(Suglia et al. 2023) [36]	Y	Y						U	Y	Y	U	Y											60%
(Chan et al. 2020) [37]	Y	Y						U	Y	Y	U	Y											60%
(El Marhraoui et al. 2025) [38]	Y	Y						U	Y	Y	U	Y											60%
(Michmizos & Krebs, 2017) [39]	Y	Y						U	Y	Y	U	Y											60%
(Belschner et al. 2024) [40]	Y	Y						U	Y	Y	N	Y											60%
(Mohammadi et al. 2024) [41]	Y	Y						N	Y	Y	N	Y											60%
(Patane et al. 2017) [42]	Y	Y						N	Y	Y	N	Y											60%
Quantitative descriptive studies
(Sobrepera et al. 2021) [43]	Y	Y											Y	U	Y	U	Y						60%
(Airaksinen et al. 2020) [44]	Y	Y											Y	U	Y	U	Y						60%
(Udayagiri et al. 2024) [45]	Y	Y											Y	N	Y	U	Y						60%
(Li et al. 2021) [46]	Y	Y											Y	U	Y	U	Y						60%
(Den Hartog et al. 2022) [47]	Y	Y											Y	U	Y	U	Y						60%
(Aguilar-Herrera et al. 2025) [48]	Y	Y											Y	N	Y	U	U						40%
Mixed methods studeis
(Tommy et al. 2024) [49]	Y	Y																Y	Y	Y	U	Y	80%
(Ostojic et al. 2022) [50]	Y	Y																Y	Y	Y	U	Y	80%
(Ng et al. 2025) [51]	Y	Y																Y	Y	Y	U	Y	80%
(Ren, 2025) [52]	Y	Y																Y	Y	Y	U	Y	80%
(Jafari et al. 2018) [53]	Y	Y																Y	Y	Y	U	U	60%

In Table 2, the appraisal outcomes are indicated using three response categories: ‘Y’ (Yes), ‘N’ (No), and ‘U’ (Can’t tell). Overall, most studies met several core MMAT methodological criteria across their respective study categories. Common strengths included the appropriateness of measurements, intervention implementation, data collection procedures, and integration of mixed-methods components where applicable.

However, several studies demonstrated limitations related to participant representativeness, insufficient reporting of confounding factor management, and limited clarity in methodological reporting, particularly among pilot and feasibility studies involving small sample sizes. In mixed-methods studies, uncertainty was also observed regarding the handling of inconsistencies between qualitative and quantitative findings.

Despite these limitations, all 23 studies satisfied the predefined inclusion criteria and demonstrated sufficient methodological rigor and relevance for inclusion in this systematic review. The MMAT appraisal supports cautious interpretation of the included evidence while also highlighting methodological areas requiring further improvement in future research.

3.2. Data extraction and analysis

Data extraction, coding, and analysis were conducted using a structured, multi-stage process designed to enhance transparency and replicability. After applying the predefined inclusion and exclusion criteria, all remaining articles underwent a quality assessment to ensure methodological rigor. As shown in Figure 1, 23 studies met the criteria and were retained for full review. Key study attributes – such as thematic focus, objectives, design/sample, research methods, AI techniques employed, outcome measures, and limitations – were systematically extracted using a standardized coding sheet developed for this review. Two reviewers independently coded the studies, and discrepancies were resolved through discussion to ensure consistency. The coded data were then analyzed using an inductive thematic approach to identify recurring patterns and conceptual groupings across studies. Based on the reviewed articles, five main themes were identified and categorized as follows: (a) AI-Driven Motor Rehabilitation and Gait Training for Functional Mobility; (b) Intelligent Assessment and Monitoring Systems for Clinical Decision Support; (c) AI-Supported Communication, Social Interaction, and Intention Recognition Tools; (d) Gamified and Virtual Reality–Based Interventions to Enhance Engagement and Usability; and (e) Smart Assistive Systems Supporting Daily Living and Independent Mobility. Each theme is discussed in detail below, with chronological and technological progress highlighted to illustrate how early foundational work enabled later innovations.

3.3. Overall certainty of evidence (GRADE assessment)

To clarify the overall strength and certainty of the evidence across the included studies, the GRADE framework was applied as described in Section 2.6. The results of this assessment are presented in Table 3 in accordance with established GRADE guidelines [30]. Overall, the certainty of evidence ranged from moderate to very low across the five thematic categories. This pattern reflects the early developmental stage of many AI-assisted rehabilitation technologies, where studies are frequently exploratory and characterized by small sample sizes, heterogeneous intervention designs, and limited use of controlled clinical trials.

Table 3.

Certainty of evidence across thematic categories assessed using the GRADE framework.

Outcome/thematic category	Number of studies	Risk of bias	Inconsistency	Indirectness	Imprecision	Certainty
AI-Driven Motor Rehabilitation and Gait Training	8	Serious	Not Serious	Not Serious	Serious	⊕⊕◯◯ Low
Intelligent Assessment and Monitoring Systems	6	Not Serious	Not Serious	Not Serious	Serious	⊕⊕⊕◯ Moderate
AI-Supported Communication and Social Interaction Tools	4	Serious	Serious	Not Serious	Serious	⊕◯◯◯ Very Low
Gamified and Virtual Reality-Based Interventions	2	Serious	Serious	Not Serious	Serious	⊕◯◯◯ Very Low
Smart Assistive Systems for Daily Living and Mobility	3	Serious	Serious	Not Serious	Serious	⊕◯◯◯ Very Low

Note: Evidence was downgraded primarily due to small sample sizes, exploratory study designs, limited use of randomized controlled trials, and heterogeneity in intervention types and outcome measures.

3.4. Characters of selected studies

Twenty-three (n = 23) eligible studies published between 2015 and 2025 were included in this review after full-text screening. The selected studies represent multidisciplinary contributions across biomedical engineering, rehabilitation science, computer science, and human–computer interaction. All included articles were published in journals focusing on robotics, artificial intelligence, assistive technology, or rehabilitation engineering. The included studies demonstrated varying levels of evidence, ranging from technical feasibility and usability evaluations to limited indications of clinical effectiveness. Most studies primarily emphasized system feasibility and usability validation, whereas comparatively few assessed quantifiable clinical outcomes related to functional improvements in children with cerebral palsy. Consequently, the current body of evidence remains largely focused on early-stage system development, with relatively limited empirical evaluation of clinically meaningful rehabilitation outcomes.

3.4.1. Publication period and research trend

As illustrated in Figure 2, the number of publications among the studies included in this review has gradually increased over the past decade, particularly from 2023 onwards, with noticeable growth in Theme 1 and Theme 2, indicating a rising research focus on the application of AI and smart technologies in pediatric neurorehabilitation. Early studies emphasized prototype development and mechanical control (e.g. robotic orthoses and exoskeletons) [31,32], whereas recent studies have explored AI-driven sensing, virtual reality-based training, and user-centered usability evaluations [33,34]. This shift indicates a transition from hardware-centric innovation to human experience-oriented design and data-driven personalization.

Figure 2.

Annual Publication Trends of AI-Assisted Rehabilitation Technologies in Pediatric Cerebral Palsy by Thematic Category (2015–2025), based on the Studies Included in this Systematic Review (source: Author’s own illustration created using Keynote).

3.4.2. Geographic and technological characteristics of the articles

As shown in Figure 3, the 23 studies included in this systematic review originate from a diverse range of countries and regions, highlighting the international scope of research on AI-assisted assistive technologies for pediatric cerebral palsy. A total number of 11 studies the research consisted of developmental or validation studies, focusing on the design, prototyping, and initial algorithmic modelling of systems [31,32,34–42], such as the AI-powered modular wheelchair from Malaysia [34,43], smart-material soft exoskeleton from the USA [37,40, 42,44,45], and AI-based pain detection application from Spain [33].

Figure 3.

Geographic Distribution of Assistive Technology Research for Pediatric Cerebral Palsy based on the Studies Included in this Systematic Review (source: Author’s own illustration created using Keynote).

In addition, several articles have reported experimental or quasi-experimental trials. These studies were typically preliminary in nature, featuring pilot studies or single-case designs with small pediatric samples, often ranging from one to 30 children with cerebral palsy or other neurodevelopmental disorders. For instance, studies from Colombia and France involved single-case or small-sample testing of a social robot and serious game platform, respectively [36,46]. Quantitative methodologies remain predominant including 12 studies, with outcomes frequently centered on technical performance metrics, such as the accuracy of AI classification, gait parameters, or range of motion. These studies combined quantitative performance data with qualitative insights gathered through surveys, interviews, or usability feedback from key stakeholders, including therapists (e.g. in a US study on a telehealth robot), parents (e.g. in a Malaysian study on a pain app), and teachers (e.g. in a Portuguese study on AT for inclusive education) [31,43,45].

3.4.3. Technological orientation

According to the screened articles, the technological spectrum covered five broad categories: (1) wearable robotic orthoses and exoskeletons for mobility enhancement; (2) virtual/augmented-reality-based motor-training platforms; (3) AI-enabled sensing, monitoring, and mobile health systems; (4) upper-limb and hand-rehabilitation devices; and (5) AI-supported communication, social-robotic, and telehealth applications. Many systems incorporated machine learning algorithms for adaptive feedback, motion recognition, and pain detection. 11 of the 12 studies integrated real-time feedback to increase engagement [31,32,34–42], and more than 14 studies explicitly evaluated usability and user satisfaction [31,33–36, 43,45–52].

3.5. Theme 1: AI-driven motor rehabilitation and gait training for functional mobility

As Table 4 shows, Theme 1, a major proportion of the selected studies (eight articles) focused on AI-enhanced robotic and sensor-based systems designed to improve the daily motor performance of children with CP, including gait, balance, and lower-limb mobility. Through these eight studies, AI was applied not only to enhance training effectiveness through adaptive, task-specific motor learning but also to improve usability, making devices more comfortable, responsive, and motivating for long-term therapeutic use. The following sections illustrate the effectiveness and usability of AI-driven designs in motor performance in children with CP.

Table 4.

Studies that focused on AI-driven motor rehabilitation and gait training for functional mobility.

Author(s)/year	Thematic focus	Objectives	Design/sample	Research methods	AI component	Outcome measures assessed	Effectiveness	Limitations
Patané et al. 2017 [42]	Portable Exoskeleton Device for Lower Limb Rehabilitation in Children with Cerebral Palsy	Design and evaluate a novel wearable ankle-knee exoskeleton, WAKE-up, to assist children with cerebral palsy in improving gait and lower limb motor function, particularly providing support during foot contact and walking.	Pre-clinical trial; 4 TD + 3 CP children	Experimental studies; Preliminary clinical trials, Quantitative study.	Position control + RSEA actuators	Ankle torque assistance effective; knee partial (max lag 8.8%)	The exoskeleton provided reliable ankle torque assistance, though knee assistance remained partial due to limited motor power.	Small sample; knee motor underpowered
Aguilar-Herrera et al. 2025 [48]	A lightweight, soft exoskeleton helps improve mobility in children with cerebral palsy.	Develop a lightweight, soft exoskeleton to help children with cerebral palsy improve their mobility and quality of life, and provide an interdisciplinary practical education platform for undergraduate students.	Prototype testing; 4-year-old proxy	Mixed-method; Case studies; Experimental studies	EMG + IMU sensor fusion; artificial muscles	Device weight; battery endurance; gait economy	The lightweight soft exosuit improved gait economy and mobility support, indicating potential effectiveness for enhancing functional walking	No long-term data; prototype stage
Mohammadi et al. 2024 [41]	DE-AFO’s new ankle-foot orthosis for children with cerebral palsy	Develop a comfortable and inconspicuous electrically powered ankle-foot orthosis (DE-AFO) that utilizes dielectric elastomer artificial muscles to improve gait function in children with cerebral palsy.	Gait lab analysis; 1 CP + 7 TD	Experimental studies; Mixed-method	Finite-state machine + DEA	Assistive force during gait phases; ankle support during swing and pre-swing phases	The orthosis demonstrated substantial assistive force (69% during pre-swing and 100% during swing), supporting improved gait assistance capability	Single CP case
Suglia et al. 2023 [36]	VR-based visuomotor adaptation in locomotion	Assess gait adaptation using embodied avatar in VR	VR trial; 13 healthy children	Experimental Studies; Mixed-method	Azure Kinect + visuomotor perturbation	Visuomotor adaptation indicators; gait biomarkers	The VR system successfully identified multiple locomotor adaptation biomarkers, demonstrating effectiveness for assessing gait adaptation	Healthy cohort only
Buitrago et al. 2019 [35]	Social robot-assisted walking (NAO)	Improve independent walking via robot interaction	Case series; 1 CP child	Case study; Qualitative study; Experimental Studies	NAO robot + SMART goals	Number of independent walking steps	The robot-assisted training enabled the child to achieve 31 independent steps, suggesting improved walking performance through robotic interaction	Single case
Belschner et al. 2024 [40]	Robotic Ankle Platform with Video Games (PedBotLab)	Assess feasibility/efficacy of PedBotLab for improving ankle ROM, strength, and motor control.	Pre- and Post-test; 10 participants with static neurological impairments.	Experimental study	Robotic ankle platform integrated with custom video game software.	Ankle range of motion (ROM); plantarflexion strength; gait speed	Significant improvements were observed in dorsiflexion ROM and plantarflexion strength, although no significant change in gait speed was reported	Game variety was limited; twice-weekly sessions may be inadequate for functional changes.
Michmizos & Krebs, 2017 [39]	Pediatric Robotic Rehabilitation (pedi-Anklebot)	Develop an adaptive robotic device (pedi-Anklebot) for sensorimotor therapy in children.	Case study; Two pilot trials at different hospitals.	Case study	Adaptive algorithm for therapy difficulty, serious games for engagement.	Ankle movement control; functional walking performance	The adaptive robotic therapy improved discrete ankle motor control and functional walking, supporting enhanced sensorimotor learning	Scarcity of pediatric devices/studies; difficult to maintain cognitive engagement in children.
El Marhraoui et al. 2025 [38]	AI-driven Serious Game for Upper-Limb Rehab (CPLAY)	Develop CPLAY platform with instrumented bricks and AI for upper-limb rehabilitation.	30 children with DCD or CP.	Case study	Transformer model for activity recognition (86% multi-class accuracy).	Rehabilitation movement classification accuracy	The AI model achieved 86% multi-class classification accuracy, demonstrating effective recognition of rehabilitation movements for adaptive therapy	Limited participants; IMU bandwidth limits simultaneous brick use; variable patient profiles.

The design of early robotic systems only provided foundational effectiveness but revealed usability constraints. For example, Patané et al. showed that the WAKE-up exoskeleton could deliver reliable torque assistance at the ankle and knee [32]; however, its bulky structure, inertia, and limited correction of complex movement patterns reduced its practical usability during real-world gait tasks. These limitations have motivated a shift toward more wearable and user-friendly design. To improve the usability of AT designs, Aguilar-Herrera et al. developed the MyoStep soft exosuit by integrating EMG (electromyography) and IMU (inertial measurement units) sensors with lightweight, compliant materials to enhance comfort and usability during ankle assistance, which directly improves usability in earlier device designs. Similarly [42], Mohammadi et al. evaluated a soft ankle exoskeleton driven by Dielectric Elastomer Artificial Muscles, demonstrating a substantial assistive force (69% during pre-swing and 100% during swing) while improving quietness, portability, and suitability for daily use [38].

Several studies have linked usability features with therapeutic effectiveness by integrating real-time feedback, adaptive algorithms, and gamified interfaces to maintain engagement. Belschner et al. examined PedBotLab, a robotic ankle trainer combined with custom video games, then found significant gains in dorsiflexion ROM (range of motion) and plantarflexion strength [44]. However, unchanged gait speed indicates that improvements at the impairment level may require an additional usability-oriented design (e.g. easier transfer and more natural movement) to translate into functional mobility outcomes.

Adaptive AI is increasingly being applied to tailor therapy difficulty and improve both user experience and training intensity. Michmizos and Krebs demonstrated this with the MIT (Massachusetts Institute of Technology) pedi-Anklebot, which dynamically adjusts task difficulty based on real-time performance patterns to maintain an optimal challenge level – an important usability feature for sustaining motivation in children [37]. Other researchers have explored how AI-mediated interfaces influence students’ engagement and adherence. Buitrago et al. found that using a NAO robot as a motivational rehabilitation companion increased the number of steps taken without falls, highlighting the motivational usability benefits of socially assistive robotics [46].

Beyond mechanical systems, AI has been applied to enhance assessment and motor learning feedback. Suglia et al. used a VR-based walking task with visual perturbations to quantify visuomotor adaptation, offering a non-invasive, user-friendly method for detecting neuromotor control deficits that complements traditional gait analysis [49]. Additionally, El Marhraoui et al. demonstrated high classification accuracy (86%) using a transformer-based model in their CPLAY (Cognitive Play) system for upper-limb therapy, showing how intelligent automation can support adaptive training, more objective progress tracking, and reduced usability burdens on clinicians and caregivers [36].

Overall, the studies in this theme provide preliminary evidence of clinical effectiveness, particularly in improving ankle movement control and lower-limb strength in children with cerebral palsy. However, several studies primarily focused on evaluating the feasibility and usability of robotic rehabilitation systems, rather than demonstrating long-term functional outcomes through controlled clinical trials. Collectively, these findings indicate that AI-powered motor rehabilitation technologies are increasingly designed to balance therapeutic effectiveness – such as improved force assistance, motor learning, and assessment accuracy – with usability considerations, including comfort, adaptability, engagement, and reduced device burden. This balance is essential for supporting sustainable functional mobility improvements and facilitating the practical adoption of these technologies in pediatric rehabilitation settings.

Theme 2: intelligent assessment and monitoring systems for clinical decision support

As shown in Table 5, Theme 2 focuses on AI technologies to monitor and measure symptoms in children with CP, providing more accurate data than traditional clinical evaluations. Across these six studies, the effectiveness of these AI monitoring systems was applied and verified through their technical precision in detecting and classifying specific conditions in children with CP during daily testing. This is discussed in the following paragraphs.

Table 5.

Studies that focused on intelligent assessment and monitoring systems for clinical Decision support.

Author(s)/year	Thematic focus	Objectives	Design/sample	Research methods	AI component	Outcome measures assessed	Effectiveness	Limitations
Den Hartog et al. 2022 [47]	Home-Based Dystonia Assessment	Investigate feasibility of home-based inertial sensors and ML to assess dystonia severity.	Experimental; 12 children/adolescents with dyskinetic CP.	Experimental	Machine Learning models (SVM, k-NN, Ensemble, etc.) for automatic dystonia scoring.	Dystonia severity scores predicted using machine learning models (F1-score)	Individually trained models achieved F1-scores of approximately 0.67–0.68, demonstrating feasibility and moderate accuracy for automated dystonia assessment	Small sample size; data from fixed moments only; generalized models performed poorly.
Airaksinen et al. 2020 [44]	Infant Posture & Movement Tracking	Develop a smart jumpsuit and ML algorithm for automatic detection of infant posture and movement.	Experimental; 22 typically developing infants (∼7 months).	Experimental	Deep CNN classifier for automatic posture and movement classification.	Posture classification accuracy; infant movement detection performance	The deep CNN classifier achieved human-level accuracy in posture and movement classification, demonstrating effective automated monitoring of infant motor behavior	Limited by situational factors (lab availability); omitted fine motor movements.
Udayagiri et al. 2024 [45]	Smart Toy for Infant Interaction Classification	Develop a smart toy using optical force sensors and ML to classify infant-toy interactions.	Experimental; 15 adults performing 2,480 interactions.	Experimental	ResNet50 model trained to classify touch, punch, weak/strong grasp.	Classification accuracy of infant–toy interaction types (touch, punch, weak/strong grasp)	The ResNet50-based system achieved high classification accuracy using six sensors, demonstrating the feasibility of automated interaction monitoring for early motor development assessment	Data collected from adults, not infants; sensor performance inconsistencies; noise in data.
Sabater-Gárriz et al. 2025 [32]	AI Mobile App for Pain Identification	Evaluate feasibility of an AI-based mobile app for pain identification in non-communicative individuals with CP.	Literature review, software planning, stakeholder consultation.	Literature review	AI-based facial recognition for pain detection from video.	System feasibility; functional requirements for AI-based pain detection	The study confirmed conceptual feasibility of AI-based facial recognition for pain identification, though the application has not yet been implemented or clinically validated	AI model performance limited by CP-PAIN dataset size; app not yet implemented or validated.
Ng et al. 2025 [51]	Mobile App for Pain Monitoring (NeuroPAIN)	Evaluate usefulness of NeuroPAIN app for pain recognition and monitoring by parents.	Prospective cohort study; 60 parents of children with bilateral CP.	Prospective cohort study	Mobile application for structured pain logging and visualization (R-FLACC score).	Pain reporting frequency using the R-FLACC score; user usability feedback from parents	The NeuroPAIN application demonstrated high usability among parents (77% reporting ease of use) and enabled systematic monitoring of pain symptoms in children with CP	Small sample; Android-only; reduced app use for children with prolonged pain episodes.
Kent et al. 2021 [31]	Systematic Process for Sensor Toolkit Development	Describe a systematic approach for user engagement and sensor evaluation for a Digital Toolkit.	Framework development; Literature review, ICF-CY framework application.	Literature review	Framework for selecting and evaluating sensor technologies for movement assessment.	Framework development for selecting and evaluating sensor technologies for mobility assessment	The study proposed a structured multidisciplinary framework for developing sensor-based assessment tools, supporting systematic technology selection for clinical applications	Focused only on mobility devices; sensor evaluation was expert-based, not empirical with end-users.

A landmark study by Den Hartog et al. demonstrated the feasibility of using inertial sensors and machine learning to assess dystonia in children with dyskinetic CP at their homes [53].

For infants, Airaksinen et al. developed a ‘smart jumpsuit’ with integrated sensors and a deep CNN (convolutional neural network) classifier that could automatically track and classify infant postures and movements with human-level accuracy, thus offering a scalable solution for early neurodevelopmental screening [39]. Similarly, Udayagiri et al. created a smart toy with optical force sensors and a ResNet50 model to classify infant-toy interactions, such as touch and grasp, to provide an automated tool for the early detection of motor delays [40].

Pain assessment is a critical challenge in this patient population. Sabater-Gárriz et al. laid the groundwork for an AI-based mobile application that uses facial recognition to identify pain and define key user requirements through a multidisciplinary process [33]. Complementing this, Ng et al. evaluated the NeuroPAIN app, a tool that allowed parents to log pain symptoms, and found that it was easy to use and helped identify that 95% of children in their cohort experienced pain, with those using assisted tube feeding at higher risk [43]. Kent et al. provided a crucial meta-perspective, describing a systematic multidisciplinary process grounded in the ICF-CY (International Classification of Functioning, Disability and Health Children and Youth Version) framework for selecting and evaluating sensor technologies, thereby emphasizing the importance of aligning technical development with clinical needs from the outset [47].

In conclusion, most studies in this theme primarily focused on the technical feasibility and accuracy of AI-based monitoring systems, with limited evidence regarding their direct impact on clinical rehabilitation outcomes. Nevertheless, these intelligent assessment technologies demonstrate strong potential for supporting data-driven and ecologically valid pediatric care, particularly through their ability to quantify dystonia, classify infant movements with human-level accuracy, and provide objective tools for pain detection and monitoring. Despite these promising developments, usability considerations remain underexplored, and the successful adoption of these systems will depend on user-centered designs that support home-based monitoring, intuitive interfaces for parents and clinicians, and frameworks that ensure meaningful clinical integration.

3.6. Theme 3: AI-supported communication, social interaction, and intention recognition tools

As shown in Table 6, Theme 3 encompasses AI-powered communication and social interaction technologies for children with cerebral palsy. These systems aim to bypass motor and speech impairments to enable more natural communication, meaningful social interaction, and greater control over the environment, ultimately reducing isolation and increasing participation in daily activities of patients. Across these four studies, various intelligent interfaces were evaluated, including speech recognition systems tailored to dysarthric language patterns, gesture- and gaze-controlled communication platforms, and socially assistive robots that facilitate structured social engagement. Together, this theme underscores the potential of AI-driven communication aids to overcome long-standing barriers in expressive and receptive communication and promote inclusion, autonomy, and quality of life for children living with cerebral palsy.

Table 6.

Studies that focused on AI-Supported communication, social interaction, and intention Recognition tools.

Author(s)/year	Thematic focus	Objectives	Design/sample	Research methods	AI component	Outcome measures assessed	Effectiveness	Limitations
Encarnação et al. 2016 [34]	Assistive Robots for Inclusive Education	Develop/test IAMCATs to enable children with disabilities to manipulate items and communicate in class.	Case study; 9 children with disabilities, 9 regular and 9 special ed teachers.	Case study	Integrated Augmentative Manipulation & Communication Assistive Technologies (IAMCATs): robot + speech-generating device.	Classroom interaction dynamics; usability of robotic manipulation and communication system	The IAMCATs system enabled children with disabilities to manipulate objects and communicate within classroom activities, demonstrating improved participation and inclusion	Small participants; physical system required more technical support.
Hsieh et al. 2023 [33]	Eye-Gaze Assistive Technology (EGAT) for Communication	Investigate impacts of EGAT on dyadic communicative interaction vs. non-EGAT condition.	Observational; 6 dyads (children/youths with complex needs & partners).	Case study	Eye-Gaze Assistive Technology (EGAT) for communication.	Frequency of communication initiation; information sharing during interaction; conversational turn-taking	The use of EGAT resulted in more frequent communication initiation and more balanced conversational interaction, indicating improved communication engagement	Small sample size; varying motor/cognitive impairments; EGAT use limited to 3–6 months.
Chan et al. 2019 [37]	Context-Aware AAC System	Design/implement/evaluate a BLE-based context-aware AAC system for children with ID.	Case study; Piloted in a special education school in Hong Kong.	Case study	Bluetooth Low Energy (BLE) beacons for context-aware Augmentative and Alternative Communication (AAC).	Response time and accuracy of AAC communication responses	The context-aware AAC system significantly improved response time and communication accuracy, enabling more spontaneous interactions for children with communication impairments	Relies on BLE beacons; small sample size (6); single school setting.
Sobrepera et al. 2021 [43]	Social Robot for Telehealth (Lil’Flo)	Evaluate therapists’ perceived usefulness of a socially assistive robot (Lil’Flo) for telehealth.	Survey research; 351 practicing therapists in the US.	Survey	Socially assistive robot (Lil’Flo) combined with telepresence and computer vision.	Therapists’ perceived usefulness of the Lil’Flo robot for telehealth rehabilitation	Therapists reported that the system improved communication, motivation, and therapy engagement, although no improvement was observed in clinical assessment functions	Non-random sampling; potential bias; video may not have effectively communicated all features.

Encarnação et al. pioneered the development of Integrated Augmentative Manipulation and Communication Assistive Technologies (IAMCATs), enabling children to control a robot arm using a speech-generating device (SGD) [31]. School teachers found it a valuable inclusion tool, although it highlighted practical challenges, such as the extra time required for activities. Eye-gaze technology is a key access method for individuals with severe physical disabilities. Hsieh et al. found that using Eye-Gaze Assistive Technology (EGAT) led to more symmetrical communication, with children initiating more interactions and providing more information [48].

Intelligence can also be embedded into the system itself, as shown by Chan et al. who used Bluetooth Low Energy (BLE) beacons to make an AAC (augmentative and alternative communication) application context aware [35]. This system automatically presents relevant vocabulary based on the user’s location, significantly improving response times and accuracy for non-verbal children. In other words, dedicated devices play a role. In the realm of social interaction, Sobrepera et al. found that therapists perceived a social robot (Lil’Flo) as useful for improving patient motivation and communication during telehealth sessions [45].

Across these studies, evidence in this theme mainly demonstrates usability and interaction improvements, while robust clinical outcome evaluations remain limited. Usability considerations such as sensor comfort, minimal calibration, automated output, and parent-friendly interfaces directly influenced system effectiveness. Collectively, this theme demonstrates the growing potential of AI-enabled assessment tools to offer accurate, continuous, and clinically meaningful information while remaining practical for real-world use by children with CP and their caregivers in clinical settings.

3.7. Theme 4: Gamified and virtual reality–based interventions to enhance engagement and usability

As shown in Table 7, Theme 4 highlights the critical role of user engagement in pediatric rehabilitation, particularly through the use of immersive and game-based strategies that make repetitive therapeutic exercises more enjoyable and motivating for children with CP. Across the four selected studies, researchers employed virtual reality (VR), augmented reality (AR), and serious gaming environments designed to simulate functional tasks, provide real-time feedback, and dynamically adjust difficulty based on user performance. These systems capitalize on the motivational appeal of interactive gaming while embedding evidence-based motor learning principles, thereby improving treatment effectiveness and usability.

Table 7.

Studies that focused on gamified and virtual reality-based interventions to enhance engagement and usability.

Author(s)/year	Thematic focus	Objectives	Design/sample	Research methods	AI component	Outcome measures assessed	Effectiveness	Limitations
Ostojic et al.2022 [50]	Biofeedback App for Pain/Anxiety (BrightHearts)	Investigate acceptability/feasibility of BART for chronic pain management in children with CP.	Mixed-methods; 10 children with CP (9–18 years).	Mixed-methods; Interview	Biofeedback-assisted relaxation training (BART) app using heart rate variability.	User acceptability; perceived benefits in pain, anxiety, sleep quality, and daily activities	Participants reported positive usability and perceived therapeutic benefits, indicating the biofeedback application may support symptom management, although no statistically significant clinical changes were observed	Small sample; no control group; some found it boring/a chore; restricted to mild intellectual impairment.
Tommy et al. 2024 [49]	Digital Therapeutic Platform with Gamification/VR	Assess current CP therapy and propose a tech-driven platform to improve engagement and outcomes.	Mixed-methods; 30 participants (15 parents, 15 teachers/therapists) from India.	qualitative questionnaire; Mixed-methods	Proposed platform with gesture control, AI-based motion tracking, gamification, and VR.	User perceptions of engagement, personalization, and therapeutic effectiveness of digital therapy platforms	Survey results indicated strong interest from parents and therapists in AI-based gamified therapy, suggesting potential effectiveness for improving engagement and adherence, though the proposed platform has not yet been implemented or clinically validated	Proposed solution, not implemented; requires long-term clinical trials; equitable access may be challenging.

Moving beyond physical motor therapy, Ostojic et al. used the BrightHearts biofeedback app, which transforms heart rate variability into a visual game, to help children manage chronic pain and anxiety [51]. Although no significant changes in pain scores were observed, children and parents qualitatively reported that it was a valuable ‘tool in their toolbox’ for managing symptoms.

Synthesizing their findings, Tommy et al. pinpointed a clear gap in traditional CP therapy, notably the lack of engagement and personalization [50]. To address this, they developed a proposal for a digital therapeutic platform that leverages VR and AI-based motion tracking. This direction was validated by their survey results, which indicated strong interest from parents and therapists and highlighted a clear demand for technological solutions to improve adherence and therapeutic outcomes.

Overall, these studies primarily demonstrate usability and engagement benefits, although evidence of measurable clinical rehabilitation outcomes remains limited. AI-enhanced gamification and VR interventions show strong combined benefits; they improve effectiveness through adaptive training and feedback while enhancing usability via motivational, engaging, and child-friendly designs that promote long-term participation.

3.8. Theme 5: Smart assistive systems supporting daily living and independent mobility

As shown in Table 8, Theme 5 focuses on intelligent assistive technologies designed to support independence in daily life for children with cerebral palsy, from manipulating objects and performing self-care tasks to ensuring safe mobility in home and community settings. These AI-enabled systems include smart wheelchairs, intelligent walkers, wearable assistive devices, and environmental control interfaces, all of which aim to reduce reliance on caregivers and enhance functional autonomy. By integrating sensor fusion, motion tracking, and real-time decision support algorithms, these tools can dynamically adapt to individual user needs and changing environmental contexts. Overall, this theme illustrates the growing potential of AI-driven assistive technologies to enhance safety and mobility and empower children with cerebral palsy to participate more fully in their daily routines and social environments.

Table 8.

Studies that focused on smart assistive systems Supporting daily Living and independent mobility.

Author(s)/year	Thematic focus	Objectives	Design/sample	Research methods	AI component	Outcome measures assessed	Effectiveness	Limitations
Li et al.2021 [46]	Smart Glove for Hand Gesture Tracking	Develop a gesture interaction framework with a smart glove for real-time hand gesture tracking.	Experimental; Validation with an optical motion capture system.		Multi-sensor data fusion using a state-switching Extended Kalman Filter (EKF).	Hand gesture tracking accuracy; real-time performance; robustness against magnetic interference	The smart glove system achieved high tracking accuracy (mean error <5°) and stable real-time performance, demonstrating effective gesture recognition for assistive interaction	Hardware is rough; lacks displacement info; lacks a refined human-computer interface.
Jafari et al.2017 [53]	Assistive Robotic System for Manual Tasks	Evaluate usability of an assistive robotic system with virtual assistance for coloring tasks.	Single-case study; 1 participant with CP.		Assistive robotic system with virtual assistance for enhancing manual performance.	Hand controllability; physical load during task performance; usability during manual activity	The assistive robotic system improved hand controllability, reduced physical effort, and enhanced ease of performing manual tasks, indicating improved functional task performanc	Single participant; focused on one specific task (coloring); long-term effects not evaluated.
Ren, 2025 [52]	AI-Powered Modular Wheelchair	Develop an AI-powered modular wheelchair to improve adaptability, usability, and safety.	Case study; User-centered design process in Malaysia.		AI-powered features, multi-modal interfaces, real-time feedback.	Design usability; adaptability; safety features of the modular wheelchair system	The user-centered design process produced a flexible modular wheelchair framework that enhances adaptability and usability, offering a practical conceptual model for future assistive mobility system	Focus on a specific region (Malaysia); design for GMFCS level 3 may not suit all; no long-term data.

For hand functions, Li et al. created a smart glove using an array of IMUs (Inertial Measurement Units) and a sophisticated sensor fusion algorithm to track 3D hand gestures with high accuracy, which can be used for both rehabilitation assessment and as a computer interface [41]. Jafari et al. evaluated an assistive robotic system that provides virtual assistance to stabilize movements during a coloring task, significantly improving hand controllability and reducing the physical load for a user with CP, demonstrating how robotics can enable participation in fine motor activities [52].

In the mobility domain, Ren adopted a user-centered design process to develop a conceptual framework for an AI-powered modular wheelchair, emphasizing adaptability, usability, and safety to meet the evolving needs of a growing child [34].

Across this theme, the evidence largely reflects feasibility studies and prototype validation, with limited clinical trials evaluating long-term functional improvements. The common emphasis is on developing AI-powered assistive technologies that enhance independence, mobility, and participation in daily activities while maintaining high usability to support seamless integration into children’s everyday environments.

4. Discussion

This systematic review synthesizes the current landscape of AI-powered assistive technologies for children with cerebral palsy (CP), revealing a vibrant state of innovation in this field. The findings demonstrate a clear paradigm shift from static, one-size-fits-all assistive devices to dynamic, adaptive, and user-centered systems. The convergence of robotics, artificial intelligence, and sensor technologies has created new possibilities for enhancing motor functions, enabling communication, and supporting independence in daily life [31–34].

A primary strength of the current research is its focus on highly personalized interventions. The use of adaptive difficulty in rehabilitation games and individualized models for dystonia assessment underscores the need to tailor technology to meet the unique and fluctuating needs of each child [37,53]. This personalization is particularly important in CP, which is characterized by substantial heterogeneity in motor abilities, cognitive functions, and rehabilitation needs. Furthermore, the growing development of home-based technologies, including sensor suits and socially assistive robots, suggests potential for expanding access to rehabilitation and continuous monitoring outside clinical settings [39,46]. Such approaches may help reduce the burden on families and healthcare systems while enabling more ecologically valid data collection.

However, the translation of these technological advancements into validated, clinically adopted tools faces several significant challenges. A recurring limitation across nearly all themes is that many studies primarily focus on technical feasibility or system usability, rather than providing robust evidence of clinical effectiveness. In particular, a large proportion of the included studies are proof-of-concept investigations with relatively small sample sizes. Several studies, especially those involving complex hardware such as exoskeletons [32,42] or social robots, have been conducted with only a limited number of participants [46]. Moreover, many studies lack control groups or comparative experimental designs, which further limits the ability to attribute observed improvements to the intervention itself. While these studies demonstrate the technological potential and usability of AI-assisted rehabilitation systems, the methodological limitations restrict the strength of conclusions regarding their clinical impact. Therefore, there is a pressing need for larger-scale, controlled, and longitudinal clinical trials that can rigorously evaluate not only feasibility and usability but also long-term clinical effectiveness, adherence, and functional outcomes in real-world rehabilitation settings. The GRADE evaluation further indicates that the overall certainty of evidence remains low to moderate, highlighting the need for larger controlled trials to strengthen the evidence base for AI-powered assistive technologies in pediatric rehabilitation.

Another critical challenge relates to technological integration and clinical utility. Several existing studies focus on a single domain – such as motor rehabilitation, communication, or functional assessment – whereas children with CP often require support across multiple domains simultaneously in their daily lives. Future research should explore integrated AI-powered systems capable of combining therapeutic exercises, communicative assistance, environmental interaction, and continuous monitoring within a unified framework. Moreover, for these technologies to achieve meaningful clinical impact, their development should be guided by robust conceptual and clinical frameworks, as suggested by Kent et al. ensuring that the technologies address clinically relevant goals and remain practical for families and healthcare professionals [47]. The high cost, technical complexity, and infrastructure requirements of many AI-based systems also remain important barriers to widespread implementation.

In addition, ethical and regulatory considerations must be carefully addressed when applying AI technologies to pediatric rehabilitation contexts. The use of AI-driven monitoring systems, wearable sensors, and data-intensive platforms raises important concerns related to data privacy, informed consent, and algorithmic transparency, particularly when dealing with vulnerable populations such as children. Ensuring compliance with medical device regulations, safeguarding sensitive health data, and maintaining transparency in AI decision-making processes are critical for building trust among clinicians, caregivers, and families.

Furthermore, issues related to socioeconomic accessibility and the digital divide should not be overlooked. While AI-powered assistive technologies offer promising opportunities for improving rehabilitation outcomes, their benefits may be unevenly distributed if access is limited by cost, technological infrastructure, or digital literacy. Families in low-resource settings may face significant barriers to adopting advanced rehabilitation technologies, including limited access to specialized equipment, internet connectivity, or professional training. Future research and policy efforts should therefore prioritize scalable, affordable, and accessible technological solutions to ensure equitable access to AI-assisted rehabilitation tools.

Finally, the issue of user-centered design remains paramount. Although several studies mention user feedback, active engagement with children with CP and their families during the design and development process is often limited. The complex and highly individualized movement patterns, as well as the diverse cognitive and communication abilities of children with CP, present unique design challenges that require iterative participatory design approaches. Engaging children, caregivers, and clinicians throughout the development process can help ensure that these technologies are not only technically effective but also comfortable, engaging, and empowering for the child, thereby increasing the likelihood of sustained adoption and real-world impact.

4.1. Implications and future research

The findings of this review have several important implications for future research. For clinicians, this evolving landscape suggests a future in which they are equipped with smarter, AI data-driven tools for assessment [49,53] and can prescribe engaging home-based therapy programs [36] that complement traditional care. For ergonomics- and design-related researchers, the priority should be on moving from feasibility studies to effectiveness and usability, especially when referring to the AI-driven motor rehabilitation product design field. There is still a lack of research on AT product design for children with CP regarding usability issues. It is important to involve children with CP and their families more actively in the production design process to ensure that their needs are met. In conclusion, future research on this topic should focus on improving the effectiveness and usability of AT products in the AI emergency age. Interdisciplinary collaboration among engineers, designers, clinicians, and caregivers should be fostered to solve complex healthcare issues [54]. This collaboration is essential to ensure that assistive technologies remain clinically relevant, up-to-date, and user-friendly for children with CP.

5. Conclusion

This systematic review examined the effectiveness and usability of AI-powered assistive technologies supporting children with cerebral palsy. Across the literature, five major themes emerged: AI-driven motor rehabilitation and gait training, intelligent assessment and monitoring systems, AI-supported communication and interaction tools, gamified and virtual-reality–based interventions, and smart assistive systems for daily living. Together, these developments highlight a shift toward assistive technologies that are adaptive, engaging, and highly personalized. AI innovations hold strong potential to enhance communication, increase independence, and support more accessible and motivating rehabilitation. Continued interdisciplinary collaboration will be essential to ensure that these technologies translate into meaningful, real-world benefits for children with CP and their families.

Funding Statement

This research did not receive any funding.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. In addition, all relevant data are included within the manuscript and its Supplementary Materials. The Supplementary Materials are publicly available on Figshare at:: https://figshare.com/s/7ffa9372e70f73911215.