Development of an AI-driven home-based rehabilitation program with real-time kinematic feedback for post-stroke upper-limb recovery using low-cost wearable sensors

Project Abstract

This study presents the development and evaluation of an AI-driven home-based rehabilitation program that provides real-time kinematic feedback for post-stroke upper-limb recovery using low-cost wearable sensors. The system integrates flexible inertial measurement units (IMUs) and surface electromyography (sEMG) sensors embedded in a lightweight wearable sleeve combined with a cloud-based AI platform to deliver personalized therapy sessions, adaptive task difficulty, and objective progress metrics. The core objective is to enhance motor relearning through task-specific, repetitive practice while ensuring accessibility, affordability, and safety in non-clinical environments. Methodologically, the research followed a multidisciplinary design that encompasses signal processing, machine learning, and rehabilitation science. Sensor data were collected from a diverse cohort of chronic and subacute stroke survivors performing standardized upper-limb tasks (reach, grasp, manipulation) and functional activities of daily living. Preprocessing included sensor fusion, drift correction, and artifact removal to generate high-fidelity kinematic and muscle activation profiles. A supervised and semi-supervised AI framework was developed to infer joint angles, movement smoothness, anticipatory postural adjustments, and effort levels in real time. The AI model was trained to recognize compensatory strategies, classify task performance, and predict potential fatigue, enabling dynamic adjustment of exercise parameters such as range of motion, tempo, and rest intervals. Real-time feedback modalities comprise augmented visual cues, haptic guidance, and auditory prompts, designed to reinforce correct movement patterns and minimize maladaptive compensations. The system provides progress dashboards for clinicians and caregivers, including adherence rates, error patterns, and outcome measures aligned with established scales (Fugl-Meyer Assessment, Wolf Motor Function Test, and Box and Blocks). A randomized cross-over trial and a longitudinal single-subject design were employed to assess efficacy, usability, and adherence over 12 weeks, with follow-up at 6 months. Primary outcomes emphasized improvements in upper-limb motor function, dexterity, and functional independence, while secondary outcomes captured user satisfaction, perceived workload, and home usability. Results indicate statistically significant gains in motor impairment reduction and functional task performance compared with conventional home exercise programs, accompanied by high adherence and favorable user experience. The AI framework demonstrated robust generalization across users and task variations, with real-time kinematic feedback reducing compensatory trunk involvement and promoting normalized movement trajectories. Importantly, participants reported increased confidence and motivation due to autotelic, goal-oriented sessions and readily interpretable feedback. Safety analyses confirmed low incidence of adverse events, with built-in thresholds to pause sessions if sensor anomalies or fatigue indicators were detected. The study contributes a scalable, cost-effective paradigm for post-stroke rehabilitation that leverages ubiquitous wearables and AI to tailor therapy, enhance neuroplastic changes, and accelerate recovery outside clinical settings. Limitations include sensor placement variability and the need for ongoing algorithm re-training to accommodate heterogeneous stroke presentations. Future work will explore multimodal integration with telehealth support, expansion to bilateral coordination tasks, and personalized long-term maintenance programs.

Project Overview

What This Project Is About

---

This project explores a home-based rehabilitation program for people recovering from a stroke, focusing on the arm and hand. It uses artificial intelligence to guide exercises and provide real-time feedback from inexpensive wearable sensors. The aim is to help users perform correct movements at home, track progress, and stay motivated.

---

---

The Problem It Addresses

---

Many stroke survivors have limited access to in-clinic therapy, which can slow recovery. Traditional rehab may be expensive or require travel. This project seeks a practical, affordable way for people to practice safely at home while receiving guidance that adapts to their needs.

---

---

Objectives of the Project

---

1. Identify a low-cost set of sensors and softwear that can track arm movements.
2. Develop an AI system that suggests personalized exercises and corrects form in real time.
3. Create a simple user interface that is easy for non-experts to use at home.
4. Test the system with stroke survivors and gather feedback on usability and effectiveness.
5. Evaluate changes in movement quality and functional ability over time.

---

---

What You Will Do Step by Step

---

Step 1: Review existing rehab methods and sensor options. Step 2: Select affordable wearable sensors and set up data collection. Step 3: Build AI models to recognize movements and provide feedback. Step 4: Design a user-friendly app interface. Step 5: Run a small user study to test usability. Step 6: Analyze data to measure improvement in motion and task performance. Step 7: Iterate based on feedback. Step 8: Document methods, results, and limitations.

---

---

Expected Outcome

A functional, cost-effective home rehab system that guides users, records progress, and demonstrates potential improvements in upper-limb function after stroke. The project should show feasibility, user acceptance, and preliminary effectiveness to justify larger studies.