A unified IoT platform that delivers neuroplasticity-driven motion mirroring, cloud-prescribed therapy, ML-based recovery prediction, and emotion-adaptive safety control in a single low-cost device family.
Five flex sensors on the healthy hand drive five independent vacuum actuators on the paralyzed hand — replicating natural movement under 3° angular error.
A wearable HR sensor enforces tiered rules — normal therapy, reduced actuator speed, or full pause — protecting cardiovascular safety at all times.
Physiotherapists prescribe per-finger angle, hold time, reps and sets through a web dashboard. Patient gloves poll Firebase and execute prescriptions on schedule.
FSR402 sensors and dual ADS1115 16-bit ADCs capture per-finger and palm contributions, achieving ±5.2% accuracy versus a Jamar dynamometer.
A 64-32-16 TensorFlow neural network classifies recovery into five stages with 95.1% accuracy, giving clinicians objective longitudinal progress tracking.
NeuroSpeed fuses VGG-Face DeepFace and a lightweight FER CNN with EAR, PERCLOS, head pose and MAR signals to drive HIGH / LOW / STOP commands within 250 ms.
A React-based dashboard for prescription, progress visualization and session review. Exercise prescription completes in 3–5 minutes per patient.
ESP32 dual-core microcontrollers, Firebase Realtime Database, Flask APIs on AWS EC2, and FastAPI WebSocket streams form a resilient four-tier system.
Built with off-the-shelf, low-cost components — total estimated hardware cost ~LKR 33,500 — making it deployable in resource-constrained rehabilitation settings.