The Wizard Apprentice: A Serious Games System in Immersive VR as a Feasible Rehabilitation Approach in Children With Cerebral Palsy

IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 33, 2025 3105 The Wizard Apprentice: A Serious Games System in Immersive VR as a Feasible Rehabilitation Approach in Children With Cerebral Palsy Cristian Camardella, Federica Serra , Caterina Linciano, Chiara Malasoma, Gabriele Carrieri , Sara Aliboni, Ilaria Bortone, Federico Posteraro, Luca Bonfiglio , and Daniele Leonardis Abstract—Virtual reality offers the opportunity to engage the participant in challenging rehabilitation exercises, proposed in the shape of serious games. Modern VR technologies can further enhance usability, allowing the participant to seamlessly interact with VR environment with bare hands, without need of external tracking systems and complex setups. In children neurorheabilitation engagement can promote motivation and attention to the exercise, two key elements for effectiveness of the rehabilitation process. In this work we developed a rehabilitation system composed of three serious games in immersive VR, with motor exercises targeting the upper limb and trunk in children with Cerebral Palsy. The participant plays in the role of a wizard apprentice, called to cast spells, to prepare potions and to ride a magic eagle. These game scenarios involve coordinated motor functions related to trajectory tracking, pick-and-place with prono-supination, and trunk balance. The presented pilot study (12 CP children, 24 sessions), focuses on the feasibility assessment of the rehabilitation method, then, it allows a more in depth analysis on the adaptation and progress of the exercise parameters through data recorded during the whole treatment. The study shows that immersive VR games are a feasible approach in rehabilitation procedures, with positive results regarding acceptability, retention, adherence to the planned exercises and absence of adverse effects in the long-term use. They also show promising results in Received 19 January 2025; revised 28 May 2025 and 4 July 2025; accepted 28 July 2025. Date of publication 4 August 2025; date of current version 12 August 2025. This work was supported by the Horizon Europe Research and Innovation Program (Social and hUman ceNtered XR-SUN Project) under Grant 101092612. (Corresponding author: Daniele Leonardis.) This work involved human subjects in its research. Approval of all ethical and experimental procedures and protocols was granted by the Meyer Hospital (Florence, Italy), under Application No. 221-2021. Cristian Camardella, Federica Serra, and Daniele Leonardis are with the Department of Excellence in Robotics and AI and the Institute of Mechanical Intelligence, Scuola Superiore Sant’Anna, 56127 Pisa, Italy (e-mail: [email protected]; [email protected]). Caterina Linciano, Gabriele Carrieri, and Luca Bonfiglio are with the University Hospital of Pisa (AOUP), 56126 Pisa, Italy. Chiara Malasoma, Sara Aliboni, and Federico Posteraro are with the Azienda USL Toscana Nord Ovest (ATNO), 56121 Pisa, Italy. Ilaria Bortone is with the Italian National Research Council (CNR), 56124 Pisa, Italy. Digital Object Identifier 10.1109/TNSRE.2025.3595420 improvements of motor functions, although a direct comparison with a control group was not included in the study. Index Terms—Rehabilitation, immersive, virtual reality, serious games, cerebral palsy, haptics, wearable. I. INTRODUCTION CHRONIC neuromotor disorders affect a wide range of pediatric neurological conditions that cause significant limitations in daily activities [1]. Among these disorders is Cerebral Palsy (CP) which comprises a group of pathologies that alter motor control and posture due to non-progressive damage to the developing central nervous system. This neurological disorder is the leading cause of motor disability in childhood (prevalence in Europe: 1.77/1,000 per year [2]) and can coincide with alterations in sensation, cognition, communication, perception and can lead to seizure disorders. CP is often characterized by impairment of the upper limbs, and therefore rehabilitation programs play a key role in improving and maximizing the function of the affected limbs and promoting independence in daily activities [3],[4]. To help people with CP achieve their functional goals, users are often subjected to conventional therapies, such as physical or occupational therapy [5],[6],[7]. While efficacy of conventional methods is well established, novel approaches show the potential to improve repeatability, quantitative measurements of patient’s progress [8],[9]. Moreover, the serious games approach can improve involvement of the user in otherwise repetitive motor tasks [10]. More recent rehabilitation approaches based on novel technologies can overcome these barriers and improve the overall rehabilitation process. The use of virtual reality (VR) exercises proposed in the shape of serious games, with or without the addition of assistive robotic devices, has emerged as one of the most explored approaches in the last years [11],[12],[13]. Virtual reality proves a powerful tool for shaping goal-oriented motor tasks, offering different advantages: flexibility and parametrization of the exercises, allowing the therapist to modulate the difficulty of the exercise for each participant, © 2025 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ 3106 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 33, 2025 the intrinsic ability of these VR systems to continuously record quantitative kinematic data, and repeatability of the treatment. These aspects together provide better conditions for fine-analysis of the participant’s progress, and therefore potential for better adaptation of the therapy to participant’s needs. Yet another very important advantage is found in the gaming approach followed by many of these virtual rehabilitation methods [14]. Gamification of the rehabilitation exercise is expected to foster participant’s engagement and her/his motivation to improve performance, which is a crucial aspect in neuro-rehabilitation, considering that active mental participation is an undeniable factor to promote brain plasticity and, thus, boost motor recovery [15]. To enhance immersion and congruence of the provided visual feedback, haptic feedback has been explored in the rehabilitation field, although typically still limited to vibratory feedback informative of hand interactions [16],[17],[18] or providing event-related postural information [19], especially in lower limb and gait rehabilitation exercises [20]. We explored in previous studies the integration of wearable haptic devices in VR rehabilitation settings [21], and we include in the presented work novel, more compact devices designed for compliance with vision-based hand tracking [22],[23]. Immersive virtual rehabilitation systems are increasingly explored in neurological rehabilitation [24]. These systems offer a twofold advantage: enhanced immersivity, which may lead to greater user engagement, and improved congruency of visual feedback, providing a direct correspondence between the user’s physical body and the virtual body representation. VR approaches tailored to pediatric populations remain relatively scarce and often fragmented. As outlined in Table I, many studies involving children with cerebral palsy (CP) have focused on short-term feasibility, proof-of-concept designs, or non-immersive desktop-based setups [25],[26], [27],[28]. Only a few studies, such as [21] and [29], have explored the potential of immersive VR in combination with wearable haptic feedback; others employed immersive VR in larger clinical groups, with exercises focused on trunk balance [30],[31]. The present study responds to these gaps by proposing and testing a fully immersive VR rehabilitation system including wearable haptic devices. The system is evaluated in a longterm, clinically supervised protocol targeting upper limb and trunk control in children with CP. This feasibility study aims to demonstrate the clinical usability, engagement potential, and capacity for quantitative tracking within an immersive, gamebased rehabilitation framework. The study further extend our previous work [21],[29] experimenting three novel serious games scenarios based on the latest VR technologies, focusing on upper-limb and trunk balance exercises. The above advances leverage usability of the system by non-technical clinical personnel, and the quality and richness of the experience proposed to the participants. The paper is structured as follows: Section II describes the three virtual scenarios and the haptic devices; Section III reports the study participants, the clinical assessment performed and a description of the parameters varied by the therapist and game metrics recorded for each virtual scenario. TABLE I SUMMARY OF SELECTED STUDIES ON IMMERSIVE OR VR-BASED REHABILITATION IN CHILDREN WITH CEREBRAL PALSY Section IV reports the results obtained and Section Vthe discussions and final conclusions. II. SYSTEM DESCRIPTION The proposed system is composed of three serious games scenarios provided to the participant by means of a Meta Quest 2 VR headset, used in stand-alone mode. The headset features embedded head and hands tracking, which is exploited by the virtual exercises implementation. Compact and lightweight haptic devices, worn at the fingertips, are used to convey tactile events in the first two upper-limb exercises. Differently, a waist-band equipped with a wireless inertial measurement unit (IMU) is used for the third trunk-balance exercise. No other tracking or wearable devices, further than the headset and either the haptic or IMU device, are needed to be worn by the patient. At the therapist side, a wide monitor is used to visualize the VR environment outside the headset. A Graphical User Interface (GUI) is also provided through a laptop PC to control the exercise status and parameters. Data measured within the VR application are sent in real-time via WiFi (UDP communication) from the headset to the host PC, where they are stored together with the GUI parameters. A. Design and Structure of the Virtual Serious Games Three virtual gaming scenarios have been developed using Unity, involving upper limb and trunk balance rehabilitation exercises. The recruited motor functions and the game logic have been developed by a mixed team of technical and clinician researchers. The first scenario includes a highly generalizable pick-and-place task with prono-supination, which is involved in many Activities of Daily Living (ADLs). The second scenario focuses on higher precision and multi-joint coordination of the upper limb. Hence, it includes a path tracking task with a final aiming and launching phase, requiring both precise upper-limb coordination to accomplish CAMARDELLA et al.: WIZARD APPRENTICE: A SERIOUS GAMES SYSTEM IN IMMERSIVE VR 3107 Fig. 1. a) Sequence of the motor task b) Task representation in the game scenario c) The haptic devices and the VR headset are worn by the child during therapy. d) Grasping phase e) Reaching phase. f) Pouring phase. g) Releasing phase. the trajectory-following task. These exercises involved one hand only (selectable between left or right). For children with hemiplegia, exercises were always played with the affected hand, while, for children with diplegia, it was played with the hand more needed to foster autonomy. The third scenario involves trunk balancing, and served also as a break inbetween upper-limb exercises. In the game setting, the child takes the role of a wizard apprentice, involved in preparing potions, launching spells, and riding a magic eagle. 1) Virtual Scenario 1: Potions Class:In the scenario the child has to learn how to prepare a magic potion, in an environment immersed in ancient books, cauldrons and magical ingredients. As part of this interactive game, the child is asked to assume the role of an apprentice magician, tasked with preparing a magical potion by mixing various ingredients in a cauldron (Fig. 1). The child is involved in pick-and-place motor tasks including prono-supination, in order to grasp the right ingredient from a shelf, and to pour the right amount of the ingredient into a cauldron. Cognitive elements are introduced to increase the child’s interest and motivation by requiring accuracy in the amount of liquid poured into the cauldron. A green reference line on a progress bar indicates the ideal amount of liquid to pour. 2) Virtual Scenario 1: Potions Class:In the scenario the child has to learn how to prepare a magic potion, within an environment immersed in ancient books, cauldrons and magical ingredients (Fig. 1). The exercise is composed of a pick-and-place motor tasks including prono-supination, in order to grasp the right ingredient from a shelf, and to pour the required amount into a cauldron. The game phases of the Potions Class scenario are: •a) Grasping: a red arrow suggests to the child the ingredient to take from among the various bottles on a table. The child grasps the ingredient by performing a pinching action with the hand. •b) Reaching: the child brings the ingredient on top of the cauldron. •c) Pouring: the child starts pouring by rotating the ingredient over the given angular threshold, and stops the pouring by rotating back the bottle below the angular threshold. A progress bar indicates the ideal amount of liquid to pour. •d) Releasing: The child places and releases the ingredient at the initial location on the shelf. To complete the potion, the cycle is repeated for five ingredients, all forming a gaming round. At the end, if the potion has been correctly prepared, a green smoke emerges from the cauldron and a magical animal appears in reward nearby the child. Otherwise, a black smoke emerges from the cauldron and no animal appears. The therapist could parameterize the exercise by varying the difficulty of the recipes, determined by the placement of the ingredients within the workspace, the speed and precision required to pour the ingredients, and the minimum pronation angle necessary to start the pouring action. Regarding haptic feedback, it conveys normal forces at fingerpads congruently to the grasping transition. During the pouring phase, a modulated vibration is added, synchronized with the sound of a flowing liquid. 3) Virtual Scenario 2: Spell Cast:In the Spell Cast scenario the child has to learn how to cast powerful spells (Fig. 2) by tracing the corresponding magical symbol. Symbols are presented in mid air one at time and have to be drawn with the index finger by following the visual trace. The drawing accuracy, compared to the shown visual reference, determines the spell’s power. After the tracing phase, the spell appears in the hand and has to be aimed and cast against an enemy, appearing in the middle of a magic circle, by following a linear trajectory. Simple cognitive elements are introduced with the aim of stimulating higher involvement in the game. In example, the suggested spells are associated with magical elements (such as Fire, Water, Wind, etc.) and in turn associated to weak points of the presented enemies. The exercise parametrization includes the difficulty of the prescribed trajectory, through a library of symbols with different complexity, the dimensional scaling of the symbol, the required tracing precision, and the moving velocity of the target enemy. 3108 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 33, 2025 Fig. 2. Spell Cast scenario and sequence of actions: a) Start tracing b) Symbol tracing c) Spell creation d) Aiming and cast e) Experimental setup. The game phases of the Spell Cast scenario are: •a) Start Tracing: the hand is either in a point or pinch pose (selectable by the therapist), with such pose held for two seconds at the starting point of the reference symbol, marked by a red circle. Then, a flame appears at the fingertip; •b) Tracing: the symbol is traced following the indicated path. •c) Spell Creation: the hand is opened, and the spell (i.e. a fire ball) is visualized floating in the hand palm; •d) Aiming: by orienting the hand, a violet ray traced from the hand palm has to be pointed towards the enemy. •e) Casting: by performing a linear movement of the hand along the violet aiming line, the spell is launched. A velocity threshold is used to release the spell from the hand. A full gaming round was composed of a sequence of three enemies to defeat. The Spell Cast environment and game sequence is depicted in Fig. 2. Haptic feedback with vibration components modulated by the tracing velocity is provided during the tracing phase. Different modulated vibrations are provided once the spell floats in the player’s hand. A 2D spatial projection and a dynamic time warping (DTW) algorithm is employed to process position tracking data in order to determine the precision score of the symbol in the game logic (Fig. 2). 4) Virtual Scenario 3: Eagle Ride:This scenario places the child on the saddle of a magic eagle, flying in a slow-paced and relaxing mountain environment. The exercise is targeted at postural balance, by means of gentle leaning of the trunk in order to guide the trajectory of the eagle. A reference path is defined by a series of waypoints Different control modalities, described hereafter, can be selected to guide the eagle using the posture of the trunk. If the player deviates from the ideal trajectory, the eagle gradually slows down. This logic makes the game robust to novice players and easier for them to find again the waypoint if it gets out of sight. Also, since the velocity of the eagle is the highest when the waypoint is correctly targeted, it promotes precision of the movement in order to indirectly increase the velocity of the game. The following control modalities were implemented: The game modalities of the Eagle Ride scenario are: •Automatic mode - Fully automatic mode, the eagle is automatically guided throughout the waypoints. The option is used the first time to get the child acquainted to the scenario. •Roll control mode - The child is required to gently lean the trunk on the right or on the left, on the sagittal plane, in order to steer the direction of flight. The steering velocity is proportional to the leaning angle of the trunk (roll axis). •Roll and Pitch control mode - Similar to the Roll control mode. In this condition, also the pitch angle of the eagle (up-down direction) has to be modulated by gently leaning the trunk upward and backward in order to reach waypoints at different heights. •Free mode - In this condition the flight trajectory of the eagle is unconstrained even if the eagle gets out of the path. Both roll and pitch angles of the trunk are used to steer the eagle left/right and up/down. Waypoints are the same than in the other conditions. Moreover, the therapist could parametrize the exercise changing the total number of waypoints in each round, and the scaling of the eagle speed. A full gaming round corresponded to the completion of single path of ten to twenty waypoints, set by the therapist. B. The Wearable Haptic Interface Tactile feedback of virtual object interaction has been included by custom developed wearable haptic thimbles. In this study the focus was to increase wearability and compliance of these devices with respect to both participants’ comfort and compliance with the embedded hand tracking system provided by the 3D headset. A light and thin haptic thimble has been developed, with direct-drive actuation targeted at rendering low amplitude, yet clean and dynamic pressure signals directly at the fingerpad tissue. These include fast transients, textures and modulated vibrations (Fig. 4). CAMARDELLA et al.: WIZARD APPRENTICE: A SERIOUS GAMES SYSTEM IN IMMERSIVE VR 3109 Fig. 3. a),b): The movements performed by the child atop the eagle to reach the bright targets are the tilting of the torso to the right or left to steer the eagle’s flight, and the trunk control to balance and maintain stability during the eagle’s flight. c) The Oculus Quest 2 is worn by the child during therapy. d),e): Display of the Eagle Ride VR scenario. Fig. 4. The thimble wearable haptic device developed for high wearability and compliance with vision-based tracking systems. a) Assembled design and b) section view of the internal voice coil. c) Detail of the soft actuated membrane in blue color. d) Device worn at the hand of a child. A miniaturized custom-made electromagnetic voice coil (outer diameter 12 mm, output force 0.4 N) was implemented due to the linear output and to the wide frequency response of this type of actuators. The thimble is fabricated in a soft resin, Photocentric Ultraviolet Digital Light Processing (UV DLP) Flexible, making it adaptable to different finger sizes. The weight of each haptic thimble is 7 g, while the weight of the wireless electronics box worn at the forearm, including the battery, is 60g The large capacity battery (1500 mAh) allows extended operation for about 4-5 hours, depending on the intensity of the haptic events. Further technical details can be found in [22]. The same electronics includes an IMU sensor (InvenSense MPU6050), and is used as a wireless inclinometer, worn at the trunk, for the Eagle Ride scenario. III. EXPERIMENTAL METHODS A. Participants Twelve children (10 male and 2 female) with Cerebral Palsy in its hemiplegic and diplegic forms were recruited for the study. The inclusion criteria were a confirmed diagnosis of CP, with participants aged between 6 and 14 years (Table II). The exclusion criteria were epileptic users, severe deficit in sensory perception of upper limb, severe visual impairments and severe cognitive involvement. All participants provided a TABLE II PARTICIPANTS minimum ability to actively use their upper limb and understand simple instructions. The parents or legal guardians of the children provided written informed consent for participation. The clinical study was approved by the pediatric ethical committee of the Meyer hospital (Florence, Italy) with protocol number 221-2021. B. Clinical Assessments Clinical assessment was performed at the beginning (Time T0) and at the end (Time T1) of the therapy, using the following scales: the Box and Block Test (BBT) [33], the Nine Hole Peg Test (NHPT) [34], the Time Up and Go (TUG) [35], the Melbourn Assessment 2 (MA2) sub-divided in the range of motion, precision, dexterity and smoothness scores. C. Measured Data The following criteria were adopted to conduct the feasibility study: •Retention - Ability to complete treatment •Adherence - Consistency and intensity of the treatment •Acceptability - Acceptability and occurrence of adverse effects •Usability - Usability of the system and occurrence of technical issues. Moreover, we analyzed two data sources: the game parameters set for each session, adapted by the therapist on the 3110 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 33, 2025 TABLE III PARAMETERS SET BY THE THERAPIST AND MEASURED METRICS basis of the progress of the participant, and the game metrics, which include kinematic data and other exercise-specific measurements. Parameters and metrics for each scenario are described in Table III. In the plotted results the parameters are normalized in the range 0 - 1, corresponding to the minimum and maximum value ever set in any session and for any participant. A 5 points Likert scale questionnaire was collected after the treatment from parents of the children. Table IV shows all the questions, related to acceptability of the proposed therapy. D. Statistical Analysis The objectives of the statistical analysis were: a) to assess a statistical significant difference on clinical scale scores, between t0and t1; b) assess a statistical difference between game parameters set by therapist between t0and t1; c) assess a statistical difference between game performance indexes between t0and t1. Given the limited sample size and the absence of normality, non-parametric tests only were run. TABLE IV ACCEPTABILITY QUESTIONNAIRE Wilcoxon signed rank paired tests were used to assess the difference among the two conditions, being the two groups dependent variables. The significance levels on the p-value have been set to 0.05, 0.01, and 0.001 corresponding to one, two, or three asterisks in the following plots. IV. EXPERIMENTAL RESULTS Measurement of the feasibility criteria obtained the following results: CAMARDELLA et al.: WIZARD APPRENTICE: A SERIOUS GAMES SYSTEM IN IMMERSIVE VR 3111 TABLE V AVERAGE GAMING ROUNDS PER REHABILITATION SESSION Fig. 5. Boxplots illustrating the distribution of responses to the six Likert scale items listed in Table IV. Retention - All the patients were able to complete the proposed protocol. Adherence - The administered exercise intensity, reported in Table V, showed balanced intensity for the three scenarios. Considering the total number of gaming rounds per day, the measured value is 9.3±1.4, showing a relatively limited intersubject deviation from the average. Each gaming round refer to the completion of either one potions, (duration of about 4 minutes). This should not be mistaken for the exercise sessions per day, which refers to the total number of all game instances. Usability - The system could be operated by clinical personnel alone, with technical personnel present during the first rehabilitation session of each new patient. Technical interventions during the therapy were limited to the following: extensions of parameters’ range on the basis of the first day of therapy; added an optional, different starting hand pose for casting spells; added the free-flight condition, added additional rewards in the Potions Class scenario, added new spells and enemies in the Spell Cast scenario. Minor interventions were performed throughout the therapy (less than 10 for hospital) related to network connection issues, setting of the workspace, software updates required by the headset. Acceptability - The responses to the Likert scale questionnaire, collected after the treatment from parents of the children, reported noticeably high scores to all the questions as shown in Figure 5). No adverse effects were recorded during the treatment. Results of the assessment using clinical scales at the beginning and at the end of the treatment is reported in Fig.6. All the adopted scales show a trend of the median value according to an improvement of motor outcomes at the end of the therapy. Significant differences are found for the BBT score (mean Start: 19.0, End: 24.4, p<0.01), for the MA2 ROM (mean Start: 0.72, End: 0.80, p<0.01), for the MA2 Dexterity (mean Start: 0.65 End: 0.74, p<0.01) and for the MA2 Smoothness (mean Start: 0.71 End: 0.78, p<0.01). Regarding the NHPT, 4 participants were not able to accomplish the test. Similarly 2 patients were not able to accomplish the TUG test. These participants were excluded from the statistical analysis of the corresponding clinical scale. Game parameters results are reported for each scenario using barplots comparing the initial set of values, chosen by the therapist for each participant at the beginning of the therapy, and the last set of values set at the end of the therapy. Game metrics results are plotted over the percentage of accomplished rehabilitation treatment (considering the number of performed sessions rather than date and time) for each children, taking into account minor discrepancies from the total number of sessions. Barplots in Fig. 7show results of the parameters set by the therapist for the Potions Scenario, normalized in the 0-1 range. All the parameters show a trend of increasing difficulty between the start and the end of the therapy, except the Pouring Angle threshold holding the same mean value. The Pouring Velocity shows a marked difference with statistical significance (median value Start: 0.06, End: 0.74, p<0.01). Regarding the measured metrics in the Potions scenario (Fig. 7), the Completion Time shows a difference with statistical significance (mean Start: 21.6, End: 14.17, p<0.01). The Completion Time and the Fallen Objects metrics show an evident trend according to an improvement in performance during therapy. Both metrics show a similar oscillatory behavior; such similarity can be explained by the relation between the two metrics, with a higher number of grasping attempts corresponding to a higher Completion Time. The Precision Score shows only a slight, yet not marked improvement (mean Start: 0.98, End: 1.03). The Hand Rotation Angle also does not show a noticeable variation throughout the therapy (mean Start: 65.5 deg End: 64.3 deg). This is in agreement with the almost constant parameter set for the pouring angle, close to 80 deg, and considering that the mean value of the measured angle includes also the transient phases of the whole prono-supination action. For the Spells Scenario, the serious game parameters shown in Fig. 8result in a noticeable variation for the Symbol Complexity Level (median Start: 0.15, End: 0.97, p<0.001), and for the Target Velocity (median Start: 0.0, End: 0.71, p< 0.05). The Symbol Dimension did not show a change from the median value (0.5, corresponding to 0.5 meters). Regarding the measured metrics in the Spells Scenario (Fig. 8), the Tracing Velocity did not show marked variations throughout the therapy (mean Start: 0.91, End: 1.04). The normalized Smoothness metric, computed on the traced trajectory, shows a significant difference (mean Start: 0.91 End: 1.16, p<0.05). Metrics of the aiming and launching phases show again a constant trend for the Aiming Time (mean Start: 11.14 s, End: 10.3 s), and a slight, not significant decrease of the Aiming Error (mean Start: 19.78 deg, End: 18.74 deg). Results of the above measured metrics should be considered together with the increased difficulty of the exercises, shown by the barplot in the same Fig. 8, and discussed more in details in the next section. Finally, parameters recorded for the the Eagle Ride Scenario are depicted in Fig. 9. Significant increments from the start to the final settings are found for the Speed Level (median Start: 0.88, End: 1, p<0.05) and for the Pitch Enabled parameters (median Start: 0.0, End: 0.5, p<0.05). The Free Flight modality changed from 0.17 to 0.42, with 5 over 3112 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 33, 2025 Fig. 6. Summary of clinical scale scores for all the children with cerebral palsy. Each boxplot shows the distribution of each variable, having the whiskers that show the 1st and 4th quartiles, the blue box that shows the 2nd and 3rd quartile, and the red line that shows the median. Red crosses are outliers. One, two or three black asterisks show the significance level with p-value respectively lower than 0.05, 0.01, and 0.001. All adopted clinical scales show a median value trend consistent with motor improvement at the end of the therapy. Fig. 7. Results of the Potions Scenario. Boxplot shows the starting and ending values of parameters set by the therapist, ranging from 0 to 1; higher values correspond to increased task difficulty. The progress plots show the average metrics over the percentage of completion of the therapy. Performance improvements correspond to higher values, except for the Fallen Objects and Completion Time, which improve with decreasing values. 12 participants experiencing the more advanced exercise feature at the end of the treatment. Measured metrics show a marked difference with statistical significance for the Heading Error (mean Start: 26 deg, End: 16 deg, p<0.05) and for the Eagle Velocity (normalized mean value Start: 0.75, End: 1.18, p<0.05). Fig. 8. Results of the Spells Scenario. Boxplot shows the starting and ending values of parameters set by the therapist, ranging from 0 to 1; higher values correspond to increased task difficulty. The progress plots show the average variation of the recorded metrics with respect to the percentage of completion of the therapy. Performance improvement corresponds to an increase of the metric values, except for Aiming Time and Aiming Error, which improve as their values decrease. V. DISCUSSIONS A. Feasibility Assessment The aim of this study was to experiment the feasibility of an immersive VR rehabilitation setup in a prolonged clinical rehabilitation treatment. Regarding the retention criteria, all the children were able to complete the rehabilitation treatment, CAMARDELLA et al.: WIZARD APPRENTICE: A SERIOUS GAMES SYSTEM IN IMMERSIVE VR 3113 Fig. 9. Results of the Eagle Scenario. Boxplot shows the starting and ending values of parameters set by the therapist, ranging from 0 to 1; higher values correspond to increased task difficulty. The progress plots show the average variation of the recorded metrics with respect to the percentage of completion of the therapy. Performance improvement is reflected in a decrease in Heading Error and an increase in Velocity. supporting the adopted protocol and the inclusion criteria. Only one child declined to participate to the study, and refused to try the VR headset due to tactile hypersensitivity in the head area. Positive results were recorded in terms of adherence to the therapy intensity, showing a limited inter-subject variation from the average, and showing the adaptability of the exercise difficulty to the different patients. Acceptability was notably high, with no adverse effects recorded throughout the therapy. We highlight that the use of the system was cautious, with numerous short breaks (1-2 minutes every 5-10 minutes of exercise) performed without the headset. The serious games approach obtained engagement of the participants, as noted by direct therapist observations. The final acceptability questionnaire, answered by parents, reported high scores for all the participants. This result has to be cautiously considered, taking into account the positive expectations that a novel medical technology might arise in the general public, which might have boosted the responses. Still, the results support the acceptability of similar procedures by the end users. Usability of the virtual scenarios was positively validated, with the clinical staff operating the system independently throughout the therapy. Technical staff supervised only the first rehabilitation session for each new participant and intervened occasionally for the technical issues enlisted in the Results section. Remarkably, one of the key advantages of the recent VR headsets is the integrated tracking system, which significantly simplifies the calibration process and improves robustness of the setup. Parametrization of the exercises was a crucial and demanding phase of the system development, requiring continuous interactions between developers and tests with clinical personnel. However, not all the parameters exposed in the GUI were varied in the end as expected, suggesting that a more refined selection could enhance system usability. As regards the haptic feedback, the study shows it can be introduced without affecting usability and acceptability of the overall system. The thimble prototypes proved robust to a relatively high number of sessions, although the soft body of the thimbles was replaced once for device. Haptic feedback contributed to the overall immersion and congruency of the sensory feedback provided to the participants, still its specific role cannot be evaluated given the presented experimental setup. B. Clinical Observations The statistically significant result obtained in the BBT indicates an improvement in the manual skills of cluster grasping, holding, transferring, and releasing objects, all useful skills for carrying out many activities of daily living. This improvement is consistent with tasks that were specifically trained by gaming scenarios and that required both proximal and distal upper limb movements, but did not involve finer digital manipulation skills (such as pincer grasp). Previous studies have defined the MCID for the BBT in children with CP as 6-7 blocks [33], hence the observed increase of approximately 8 blocks indicates clinical relevance of the improvements in gross manual dexterity. Regarding the Melbourne Assessment 2 (MA2), our study showed significant improvements in range of motion, dexterity, and smoothness. These domains correspond closely to upper limb coordination and movement quality, outcomes that have also been reported in prior VR-based interventions, including those using desktop systems or semi-immersive feedback [11], [12]. According to Wang et al. [36], the MCID thresholds for these subdomains are 2.35 (ROM), 2.22 (Dexterity), and 3.20 (Fluency/Smoothness). The magnitude of score shifts observed in our study meets or surpasses these thresholds, suggesting that change are clinically meaningful. In contrast, the Precision MA2 parameter and the Nine Hole Peg Test (NHPT), which assesses fine finger dexterity, did not change significantly. This is consistent with prior studies reporting limited NHPT gains unless the intervention specifically targets digital motor control. Still, the average NHPT improvement was in line with the MCID reported by Carey et al. on the same population [37]. Regarding the trunk balance exercise, the discrepancy between the results of the TUG and the Heading Error and Velocity metrics of Eagle Ride Scenario is likely to be found in the fact that in the evaluation of the former, dynamics of walking and postural transitions have a meaningful weight, thus providing an indirect and non-specific evaluation of balance, whereas metrics relating to Eagle Scenario provide a measure of the skills specifically trained in the game (i.e., the ability to control trunk movements and to distribute body weight in the static upright position) [36],[38]. Although not significant, a reduction in TUG time from approximately 10 seconds to 6 seconds suggests a meaningful gain, given that the MCID for TUG ranges from 0.2 to 5.3 seconds depending on the GMFCS level.