Automated Gait Analysis and Assistive Device Optimization via AI-Driven Ultrasonic Sensor Fusion for Post-Stroke Rehabilitation

Project Abstract

This study presents an integrated framework for automated gait analysis and personalized assistive device optimization for post-stroke rehabilitation through AI-driven fusion of ultrasonic sensing data. The core objective is to develop a non-invasive, cost-effective, and scalable pipeline that accurately quantifies gait parameters, identifies aberrant movement patterns, and dynamically adapts assistive device settings to maximize motor recovery and functional mobility. We combine multi-sensor ultrasonic data with advanced machine learning models to capture spatiotemporal foot-ground interactions, joint angles, limb trajectories, and propulsion dynamics in real time. A novel sensor fusion architecture leverages calibrated ultrasonic time-of-flight measurements, Doppler-derived flow-like motion cues, and probabilistic spatial mapping to reconstruct high-fidelity gait kinematics under diverse rehabilitation scenarios, including varied walking speeds, terrains, and user-specific compensatory strategies. Methodologically, the project deploys a hierarchical AI system consisting of (i) a signal pre-processing and denoising module tailored to ultrasonic modalities, (ii) a multi-view fusion network that integrates temporal sequences and spatial priors to generate robust gait feature embeddings, and (iii) a decision layer that translates features into actionable device parameter adjustments (e.g., orthotic stiffness, assist-as-needed torque, and feedback intensities) using reinforcement learning and constraint-based optimization. The system’s adaptability is enhanced by personalized calibration routines that align model priors with individual patient morphology, impairment level, and locomotor goals. To validate clinical applicability, we conduct a multi-phase evaluation including retrospective data analysis, prospective healthy-subject benchmarks, and a pilot study with post-stroke participants under supervision. Key outcomes focus on (a) accuracy and repeatability of computed gait metrics such as cadence, step length, symmetry indices, and energy expenditure proxies, (b) sensitivity to subtle rehabilitative improvements over conventional assessment methods, (c) responsiveness of device parameter tuning to real-time gait perturbations, and (d) user comfort, safety, and adherence in daily living contexts. The evaluation framework incorporates cross-validation across cohorts, ablation studies to quantify the contribution of each sensor modality, and robustness tests against sensor misalignment and environmental noise. Anticipated contributions include a scalable, low-cost gait assessment tool that provides objective metrics and closed-loop device optimization, a dataset of synchronized ultrasonic gait signals for post-stroke populations, and clinically translatable guidelines for integrating AI-driven sensor fusion into existing rehabilitation protocols. The study aims to bridge the gap between quantitative gait analytics and practical, personalized device therapy, thereby accelerating motor recovery trajectories and enhancing long-term functional independence for stroke survivors.

Project Overview

What This Project Is About

---

A plain-language overview of how gait data can be measured and used to improve prosthetics and walking aids after a stroke, using simple sensors and AI to interpret movement in real time. The project tests a system that collects movement information from ultrasonic sensors, analyzes it with lightweight AI, and shows how devices can be tuned to help people walk more confidently and safely. It focuses on making gait improvements practical for clinical use and at-home monitoring.

---

---

The Problem It Addresses

---

Many stroke survivors struggle with uneven step patterns and balance, which can slow recovery and raise fall risk. Traditional gait analysis requires bulky lab equipment and expert interpretation. This project addresses the gap by offering a compact, low-cost approach that provides immediate feedback to clinicians and users, helping tailor assistive devices to individual walking patterns.

---

---

Objectives of the Project

---

1. Identify key gait metrics that reflect post-stroke walking challenges.
2. Develop a lightweight ultrasonic sensor setup to capture leg motion data.
3. Train a simple AI model to interpret gait signals and suggest device adjustments.
4. Prototype an interface for clinicians and users to view feedback.
5. Evaluate accuracy and usability with volunteers in a controlled setting.

---

---

What You Will Do Step by Step

---

1) Review basic gait concepts and select measurable indicators.

2) Design and assemble the ultrasonic sensor array and data pipeline.

3) Collect walking data from participants under different assistive settings.

4) Train and test an AI model to map sensor data to device adjustments.

5) Build a user-friendly interface to show recommendations and trends.

---

---

Expected Outcome

A functioning demonstration showing how sensor data can guide personalized assistive-device tuning, with initial evidence of improved gait consistency and safety indicators for post-stroke users.