Development of an AI-powered assistive robotics gait-training system for post-stroke motor rehabilitation using real-time kinematic feedback and patient-specific adaptation

Project Abstract

This study presents the design, development, and evaluation of an AI-powered assistive robotics gait-training system for post-stroke motor rehabilitation, integrated with real-time kinematic feedback and patient-specific adaptation. The system combines a wearable robotic exoskeleton with an embedded AI controller that leverages motion capture data, force sensing, and electromyography to dynamically adjust assistance levels for individual gait impairments. A multi-layered approach is employed (i) real-time gait phase detection and anomaly detection to ensure safe and targeted support, (ii) patient-specific modeling using supervised and reinforcement learning to capture post-stroke motor intentions, compensatory strategies, and recovery trajectories, and (iii) adaptive feedback mechanisms that guide neuroplastic changes through synchronized sensory cues (haptic, visual, and auditory) aligned with therapeutic goals. The methodology encompasses hardware prototyping, software architecture, and rigorous validation using a two-phase experimental design with healthy controls for safety verification and a pilot cohort of post-stroke individuals at subacute and chronic stages. Primary outcomes include improvements in gait symmetry, velocity, and asymmetry indices, measured via inertial measurement units and pressure-sensor insoles, as well as clinically validated scales such as the Fugl-Meyer Assessment for Lower Extremity (FMA-LE) and Timed Up and Go (TUG). Secondary outcomes assess neurophysiological correlates of motor relearning through changes in muscle activation patterns (via surface EMG) and cortical engagement inferred from portable EEG markers during training sessions. The AI components comprise a hybrid model that integrates a recurrent neural network for temporal sequence prediction with a Bayesian optimization module to personalize assistance profiles, ensuring safe exploration of motor strategies and preventing patient frustration. The control framework emphasizes impedance-based torque modulation and trajectory shaping to foster active participation while maintaining joint integrity. Data fusion techniques handle multimodal inputs, smoothing noise and handling missing data to preserve robustness in clinical environments. A randomized controlled sub-study compares the proposed system with conventional treadmill-based gait therapy and with conventional assist-as-needed robotics, evaluating effectiveness, adherence, and user satisfaction. Statistical analyses include mixed-effects models to account for repeated measures, with effect sizes and clinically meaningful thresholds predefined a priori. Ethical considerations prioritize patient safety, informed consent, and data privacy, with continuous monitoring and emergency stop capabilities embedded in both hardware and software layers. The results demonstrate significant gains in real-world walking speed, step length symmetry, and functional mobility, accompanied by favorable trajectories in neurorehabilitation biomarkers, suggesting enhanced motor recovery mechanisms through targeted, patient-specific AI-driven assistance. The study discusses scalability, clinical translation pathways, cost-benefit implications, and potential refinements in artifact rejection, adaptation speed, and multimodal feedback modalities to maximize long-term rehabilitation outcomes.

Project Overview

What This Project Is About

---

This project explores a smart robotic gait trainer that helps people recovering from a stroke. It uses an AI system to adjust walking support in real time, based on how the person moves. The goal is to make training safer, more effective, and personalized to each patient’s needs.

---

---

The Problem It Addresses

---

Many stroke survivors struggle with walking again due to weak leg muscles and poor coordination. Traditional therapy can be repetitive and one-size-fits-all. The project aims to fill this gap by providing adaptive, data-driven support that responds to a person’s progress and challenges.

---

---

Objectives of the Project

---

1. Understand how gait is affected after stroke and what an assistive device needs to support safe training.
2. Develop an AI system that interprets movement data in real time.
3. Integrate the AI with a robotic gait trainer to personalize assistance.
4. Test safety, usability, and effectiveness with simulated and real user data.
5. Provide guidelines for clinicians on deploying the system.

---

---

What You Will Do Step by Step

---

1. Review relevant literature on post-stroke gait rehab and assistive robotics.
2. Define user needs and safety requirements for the device.
3. Collect movement data from volunteers (with ethics approval) and simulate scenarios.
4. Develop real-time data processing and AI adaptation algorithms.
5. Integrate AI with the gait trainer hardware and run pilot tests.
6. Analyze improvements in gait metrics and user experience.
7. Refine the system based on feedback and test robustness.
8. Prepare a practical deployment plan for clinics.

---

---

Expected Outcome

The project should deliver an AI-powered control system that tunes gait assistance in real time, a validated set of gait improvement metrics, and clear guidance for clinicians on using the device in rehabilitation programs.