Upper limb rehabilitation medical robots represent a pivotal advancement in assisting patients with motor impairments resulting from stroke, spinal cord injury, or post-surgical conditions. These systems promise to overcome the limitations of traditional manual therapy, such as therapist fatigue, inconsistent training intensity, and lack of quantitative progress tracking. However, widespread adoption is often hindered by a misalignment between technological capability and genuine user experience. Many existing devices are perceived as bulky, intimidating, and operationally complex, failing to address the nuanced needs of patients, clinicians, and maintenance personnel. This creates a significant gap between the potential and the practical utility of the rehabilitation medical robot.
This study posits that the path to effective innovation lies not in pursuing technical specifications in isolation, but in systematically translating multi-stakeholder user demands into concrete engineering design parameters. We adopt a user-centric framework, focusing on three core user groups: the patients (end-users), the treating clinicians, and the equipment maintenance technicians. By integrating the KANO model for demand classification and Quality Function Deployment (QFD) for technical translation, this research establishes a closed-loop methodology to drive the design of upper limb rehabilitation medical robots from mechanistic aids to intelligent, empathetic partners in the recovery journey.
User Research and Demand Analysis via KANO Model
The initial phase focused on comprehensively mapping the needs within the ecosystem of the medical robot. Research was conducted across two primary usage scenarios: clinical rehabilitation centers and home-based care. The participant pool was strategically sampled to include patients with various upper limb dysfunctions, rehabilitation physicians/therapists, and technical support staff. Data collection employed a mixed-methods approach, combining KANO questionnaires, semi-structured interviews, and behavioral observation.
The KANO model is instrumental in categorizing user needs based on their impact on satisfaction. It classifies attributes into five types: Must-be (M), One-dimensional (O), Attractive (A), Indifferent (I), and Reverse (R). A paired-question format for each function was used: one question measuring satisfaction if the feature is present, and another measuring dissatisfaction if it is absent.
| User Requirement (Example) | If Function is Present (Satisfaction) | If Function is Absent (Dissatisfaction) | Kano Category |
|---|---|---|---|
| Natural, unhindered joint movement during wear | It should be that way | Very dissatisfied | Must-be (M) |
| Real-time encouraging prompts during exercise | I like it | It doesn’t matter | Attractive (A) |
| Automatic adjustment of force based on my state | I expect it | Dissatisfied | One-dimensional (O) |
The collected responses were analyzed to determine the category for each requirement and to calculate quantitative indices: the Satisfaction Increment (Better) coefficient and the Dissatisfaction Decrement (Worse) coefficient. These are calculated as follows:
$$Satisfaction \ Index (SI) = \frac{(A + O)}{(A + O + M + I)}$$
$$Dissatisfaction \ Index (DI) = -\frac{(O + M)}{(A + O + M + I)}$$
Analysis of the KANO surveys from 362 valid responses revealed distinct priority layers for each user group. For patients, Must-be (M) requirements centered on safety and basic comfort: natural joint movement (high Worse coefficient), skin-friendly materials, and support for multiple training postures. One-dimensional (O) needs, whose fulfillment linearly increases satisfaction, included pain relief, improvement in daily living activities, and a friendly user interface. Attractive (A) features, which delight users but are not expected, involved social sharing functions, family participation modes, and high environmental adaptability of the medical robot.
Clinicians prioritized data intelligence and safety. Must-be needs included automatic warning of abnormal movement patterns and reliable data analysis. One-dimensional demands were focused on predictive analytics for recovery and automated report generation. Attractive features involved seamless data integration with hospital systems and tools for collaborative research.
Maintenance personnel emphasized reliability and serviceability. Must-be requirements were a clear online repair tracking system and automatic safety cut-offs. One-dimensional needs included remote diagnostic capabilities and detailed maintenance manuals. Attractive features comprised self-check reports with guided repair videos.
Translating Demands into Design: The QFD Framework
Having categorized and prioritized user demands, the next challenge was their translation into technical specifications. This was achieved by constructing a Quality Function Deployment (QFD) “House of Quality.” The left wall of the house contains the prioritized user needs (Whats), derived from the KANO analysis. The ceiling holds the technical design elements (Hows). The central relationship matrix defines the strength of the correlation between each user need and each technical element.
To create a comprehensive set of “Hows,” technical elements were derived from three core dimensions essential for a next-generation rehabilitation medical robot: Industrial Engineering (P), Interactive Emotion (E), and Intelligent Algorithms (I).
| Dimension | Code | Technical Design Element (How) |
|---|---|---|
| Industrial Engineering (P) | P1 | Joint Degrees of Freedom |
| P2 | Ergonomic Mechanical Structure | |
| P3 | Mechanical Friction Coefficient | |
| P4 | Modularity Level | |
| P5 | Battery Cycle Life | |
| P6 | Material Antibacterial Rate | |
| Interactive Emotion (E) | E1 | Motion Feedback Delay |
| E2 | Tactile Feedback Precision | |
| E3 | Level of Feedback Emotionalization | |
| E4 | Multimodal Fusion Degree | |
| E5 | Augmented Reality Guidance Efficiency | |
| Intelligent Algorithms (I) | I1 | Big Data Engine Processing Speed |
| I2 | Accuracy of DL Model Predicting Fatigue | |
| I3 | Predictive Maintenance Algorithm Efficiency | |
| I4 | AI Anomaly Detection False Alarm Rate |
Experts scored the relationship between each user need and each technical element (strong=5, medium=3, weak=1, none=0). The absolute weight (Ej) and relative weight (Ej*) for each technical element were then calculated. The absolute weight for a technical element j is the sum of the importance ratings of all related user needs, multiplied by their relationship scores:
$$E_j = \sum_{i=1}^{n} (I_i \times R_{ij})$$
where \(I_i\) is the importance of user need \(i\), and \(R_{ij}\) is the relationship score.
The relative weight is then calculated for ranking:
$$E^*_j = \frac{E_j}{\sum_{j=1}^{m} E_j} \times 10$$
The QFD analysis yielded clear priority for the medical robot’s development. Intelligent Algorithm elements, particularly the accuracy of the deep learning model for fatigue prediction (I2) and the processing speed of the data engine (I1), received the highest weights. This underscores that the core value shift is from passive motion assistance to intelligent, adaptive training. Key Industrial Engineering elements like joint degrees of freedom (P1) and ergonomic structure (P2) followed, forming the essential physical foundation for safe and biomechanically sound therapy. Interactive Emotion elements, such as multimodal fusion (E4) and AR guidance (E5), also held significant weight, highlighting their role in enhancing engagement and reducing the cognitive load for patients using the medical robot.
Design Strategies and System Implementation
Guided by the QFD output, three core design principles were formulated for the innovative medical robot: Safety-First, Personalized Adaptation, and Emotional Connection.
1. Safety-First & Personalized Adaptation via Intelligent Control: To address the high-priority needs for natural movement and adaptive training, the control system integrates a hybrid LSTM-CNN neural network architecture. The Long Short-Term Memory (LSTM) network processes time-series data from surface electromyography (sEMG) and inertial sensors to track user state and predict fatigue trends. The Convolutional Neural Network (CNN) extracts spatial features from sEMG signals to recognize movement patterns and intent. This combined model enables the robot’s impedance controller to dynamically adjust assistance/resistance levels in real-time, creating a truly “patient-in-charge,” personalized training experience that is both safe and effective.
2. Emotional Connection through Multimodal Interaction: To combat the boredom and isolation of repetitive therapy, the medical robot employs a rich, multimodal interaction layer. This includes: Augmented Reality (AR) guidance that overlays virtual tasks (e.g., reaching for a cup) onto the real world; emotionally-tuned auditory feedback (encouraging voice, adaptive soundscapes); and precise tactile feedback via linear resonant actuators (LRAs) to provide intuitive cues about movement quality or errors. A gamified progression system with rewards and visualizations of improvement directly targets the user’s psychological need for motivation and achievement.
3. System Robustness for Real-World Deployment: The reliability of the medical robot across diverse users and environments is paramount. A robustness verification framework was designed, testing the system against three categories of interference:
| Test Dimension | Scenario | Key Metric |
|---|---|---|
| User Behavior | Simulating sudden muscle spasm via sEMG signal injection | Response delay to safety mode switch ≤ 50ms |
| Environmental Interference | Operation under strong background noise (80 dB) | Voice command recognition success rate ≥ 85% |
| System Fault Tolerance | Simulated disconnection of a single motor | Graceful degradation to a safe, limited functionality mode within ≤ 100ms |
The technical implementation features a 7-degree-of-freedom hybrid serial-parallel mechanical structure for natural movement replication, using lightweight materials like titanium alloy. A multi-modal sensor suite (sEMG, 6-axis force/torque, optical motion capture) feeds data to the central AI processing unit. The software architecture is layered, separating real-time control, AI algorithm services, and the interactive application layer to ensure both responsiveness and stability.
Conclusion
This research demonstrates a systematic, user-demand-driven framework for innovating in the field of upper limb rehabilitation medical robots. By applying the KANO model, we moved beyond a simple list of wants to understand the emotional impact and priority of features for patients, clinicians, and technicians. The subsequent QFD process successfully translated these often-abstract human needs into a weighted set of technical specifications spanning industrial engineering, interactive design, and intelligent algorithms.
The resulting design strategy shifts the paradigm for the rehabilitation medical robot from a tool that merely moves limbs to an intelligent partner that understands, adapts, and motivates. It prioritizes safety through robust engineering and real-time AI monitoring, enables personalization through biosignal-informed adaptive control, and fosters engagement through empathetic, multimodal interaction. Furthermore, the proposed robustness verification methods ensure these advanced features perform reliably in the unpredictable context of clinical and home care.
This work provides a replicable methodology and a concrete design direction for developing the next generation of medical robots. The ultimate goal is to create devices that are not only technologically sophisticated but are also trusted, usable, and genuinely beneficial companions on the patient’s road to recovery, thereby bridging the critical gap between engineering potential and therapeutic reality for upper limb rehabilitation.