Project Overview

Body Machine Interfaces (BMIs) are emerging human-computer interaction systems that translate physiological signals — such as surface electromyography (sEMG), electroencephalography (EEG), and motion data — into control commands for machines, prosthetics, robotic exoskeletons, and other cyber-physical devices. As BMIs become increasingly integrated into medical, rehabilitation, and industrial settings, the security of the monitoring-control loop connecting the human body to downstream actuators becomes a critical concern.

This project investigates the attack surface of BMI systems, with a particular focus on adversarial manipulation of biosignal acquisition, transmission, and processing pipelines. We explore how physical-layer attacks, signal spoofing, and adversarial machine learning techniques can compromise BMI integrity, and develop principled defenses to safeguard patients and users from harm.

Research Objectives

  • 01
    Threat Modeling for BMI Systems

    Systematically characterize the attack surface of the monitoring-control loop, from biosignal capture through wireless transmission to actuator command execution, using established frameworks such as MITRE ATT&CK.

  • 02
    Physical-to-Cyber Attack Feasibility

    Investigate whether adversaries can exploit physical proximity or environmental interference to inject malicious signals into sEMG or EEG acquisition pipelines, causing unintended or harmful machine actions.

  • 03
    Adversarial Robustness of BMI Classifiers

    Evaluate the resilience of deep learning-based gesture and intent recognition models against adversarial perturbations, and develop robust training strategies and anomaly detection mechanisms.

  • 04
    Secure Control Loop Design

    Propose architectural countermeasures — including signal authentication, integrity verification, and safe-mode fallback mechanisms — to harden the end-to-end BMI pipeline against both cyber and physical threats.

Approach & Methodology

The project takes a co-design approach, combining hardware-in-the-loop experimentation with a custom BMI testbed, adversarial machine learning research, and formal security analysis. We use commercial off-the-shelf sEMG sensors and microcontrollers to prototype realistic attack and defense scenarios. Findings are validated through controlled experiments and evaluated against real-world usability constraints.

Expected Outcomes

The project aims to deliver a comprehensive threat taxonomy for BMI systems, a suite of open-source attack simulation tools, and a set of security guidelines and defensive mechanisms suitable for adoption by BMI device manufacturers and healthcare regulators. Results will be disseminated through peer-reviewed publications at top-tier security and human-computer interaction venues.

Members

Former Members