From sensing to actuation in a single closed loop — through embedded boards, the cloud, machine-learning inference, and a clinician-facing dashboard.
Five flex sensors on the healthy-hand glove produce analog resistance proportional to bend angle. A heart-rate sensor reads continuously, and an RGB camera captures the patient's face for landmark extraction.
Dual-core ESP32 microcontrollers sample the ADCs, convert raw values to angular degrees, run safety checks, and stream data over Wi-Fi to the cloud — all while controlling MOSFET-driven vacuum pumps on the actuation side.
Sensor frames, prescriptions, and device state synchronize through Firebase at 10 Hz with sub-200 ms latency. Patient gloves poll for active prescriptions; clinician dashboards subscribe to live progress feeds.
A Flask API on AWS EC2 receives grip-strength frames and runs a TensorFlow neural network that places each patient on a five-stage recovery scale. NeuroSpeed runs ensemble emotion recognition (DeepFace + FER CNN) at 4–10 fps.
Five vacuum-driven soft actuators replicate the prescribed bend angles independently per finger, with force kept below 5 N for clinical safety. Tiered HR rules and emotion signals can slow or pause therapy in real time.
Physiotherapists review session history, longitudinal grip recovery trajectories, emotion timelines, and adjust the next prescription — all from a React web application.
Flex sensors, FSR402 ×6, MAX30100 HR, ADS1115 ADCs, RGB camera.
ESP32 dual-core boards: ADC sampling, actuator control, Wi-Fi bridge.
Firebase RTDB for state sync, Flask + AWS EC2 for ML, FastAPI WebSocket for streams.
React clinician dashboard plus mobile companion for the patient.