Pareto-Optimal Model Selection for Low-Cost, Single-Lead EMG Control in Embedded Systems

TL;DR

This work tackles reliable hands-free EMG control using ultra-low-cost, single-lead sensors on resource-constrained edge devices. It benchmarks 18 architectures from heuristics to MaxCRNN, demonstrating that Random Forest with simple statistical features provides the Pareto-optimal embedded solution, while MaxCRNN defines the safety ceiling under higher compute budgets. A novel MaxCRNN architecture (Inception + Bi-LSTM + Attention) achieves 99% CLENCH precision and 83.21% overall accuracy on a single-subject dataset, though ESP32 deployment remains infeasible for such a model. The study provides a reproducible dataset and code, showing that reliable, low-latency EMG control is feasible on commodity hardware, with deep learning offering high reliability on modern edge accelerators.

Abstract

Consumer-grade biosensors offer a cost-effective alternative to medical-grade electromyography (EMG) systems, reducing hardware costs from thousands of dollars to approximately $13. However, these low-cost sensors introduce significant signal instability and motion artifacts. Deploying machine learning models on resource-constrained edge devices like the ESP32 presents a challenge: balancing classification accuracy with strict latency (<100ms) and memory (<320KB) constraints. Using a single-subject dataset comprising 1,540 seconds of raw data (1.54M data points, segmented into ~1,300 one-second windows), I evaluate 18 model architectures, ranging from statistical heuristics to deep transfer learning (ResNet50) and custom hybrid networks (MaxCRNN). While my custom "MaxCRNN" (Inception + Bi-LSTM + Attention) achieved the highest safety (99% Precision) and robustness, I identify Random Forest (74% accuracy) as the Pareto-optimal solution for embedded control on legacy microcontrollers. I demonstrate that reliable, low-latency EMG control is feasible on commodity hardware, with Deep Learning offering a path to near-perfect reliability on modern Edge AI accelerators.

Figures (43)

• Figure 1: Architecture Evaluation Pipeline. The system processes EMG signals through three competing logical paths: (A) Statistical Feature Engineering for ultra-low latency, (B) Transfer Learning for visual texture analysis, and (C) Custom Deep Learning (MaxCRNN) for maximum precision. The Random Forest (Top) was selected for the embedded ESP32 implementation, while the MaxCRNN (Bottom) establishes the theoretical safety ceiling.
• Figure 2: Target Muscle: Flexor Digitorum Profundus. I target this deep muscle to detect grip intent for turn signal activation.
• Figure 3: Time-domain comparisons of the three classes. Note the stochastic nature of the CLENCH signal versus the structured mechanical artifacts in NOISE.
• Figure 4: Mel-Spectrograms reveal distinctive textures. 'CLENCH' (Left) shows broadband vertical noise, while 'NOISE' (Right) exhibits frequency-specific bands.
• Figure 5: Feature Space Separation: MAV vs ZCR. Note the overlap between Noise and Clench, which simple thresholding fails to resolve.