Optimization of wrist joint kinematic decoding Based on HD-sEMG

Read full text Summarise with VIRAT
Authors: Gerard Gómez Izquierdo
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The restoration of intuitive motor function for upper-limb amputees remains a critical challenge in rehabilitation robotics. While Deep Learning (DL) has enabled robust Simultaneous and Proportional Control (SPC) of prosthetic wrists, state-of-the-art architectures typically rely on highperformance computing hardware, rendering them unsuitable for the strict power and thermal constraints of wearable embedded systems. This thesis addresses this "embedded gap" by developing a lightweight Convolutional Neural Network (CNN) capable of decoding continuous wrist Flexion/Extension and Pronation/Supination from High-Density sEMG (HD-sEMG) directly on a standard microcontroller. The proposed framework introduces a spectral-spatial feature extraction pipeline that transforms 32- channel sEMG signals into stacked Frequency-Spatial Maps, allowing the CNN to capture dynamic muscle recruitment patterns without the latency of recurrent layers. To guarantee smooth trajectory reconstruction, an adaptive filter [9] was integrated, significantly reducing signal jitter with negligible computational cost. To ensure deployment feasibility, the architecture was further optimized via 8-bit Post-Training Quantization for the ARM Cortex-M7 processor. Validation across five healthy subjects demonstrated that the quantized model retains high decoding fidelity, matching the performance of significantly larger state-of-the-art models with an R 2 ≈ 0.92 for isolated movements and variance-weighted R 2 ≈ 0.84 for complex simultaneous tasks. Crucially, hardware profiling confirms the system’s efficiency: the model occupies only 29.2 KB of memory and achieves an inference latency of 20 ms. These results confirm that state-of-the-art simultaneous prosthetic control of the wrist can be achieved in an embedded device, effectively bridging the gap between advanced EMG decoding and viable clinical hardware.

Fields of Study & Subjects

Muscle activation and electromyography studies Prosthetics and Rehabilitation Robotics Stroke Rehabilitation and Recovery

Publication details

Journal / source
UPCommons institutional repository (Universitat Politècnica de Catalunya)
Publication date
11 February 2026
Language
EN
Document type
Thesis