sEMG-Based Robot Control as an Experimental Platform for the Development of Technologies Applied to Rehabilitation

Abstract

Introduction: Human-machine interfaces based on surface electromyography allow the transformation of voluntary muscle activation into functional commands for controlling robotic devices. The integration of myoelectric sensors, embedded processing, wireless communication, and mobile robotics constitutes an accessible alternative for developing experimental prototypes geared towards functional assistance and motor rehabilitation.

Objective: To develop and calibrate a robotic control system based on surface electromyography signals, embedded processing, and wireless communication to control the navigation of a mobile robotic vehicle, as an experimental platform for technologies applied to rehabilitation.

Methodology: A modular architecture was designed, consisting of four stages: myoelectric acquisition, embedded processing, wireless communication, and robotic actuation. For acquisition, MyoWare© surface electromyography sensors were used, placed on forearm muscle groups associated with voluntary hand and wrist movements. The commands were linked to functional motor patterns: hand closing to move forward, hand opening to move backward, wrist deviation to the right and left for lateral turns, and muscle relaxation to stop the vehicle. In the processing, an ESP32 microcontroller performed analog reading, basic digital conditioning, activation threshold identification, and conversion of muscle patterns into discrete commands. Wireless communication transmitted the instructions to the robotic vehicle's controller. Initial tests were conducted using a commercial Myo Armband © gesture interface to validate the gesture-command logic. Subsequently, the system was upgraded to independent myoelectric sensors to increase the flexibility of acquisition, calibration, and control adaptation. Calibration determined the overall recognition rate, rate per command, average latency, standard deviation of latency, and operational performance.

Results: Technical calibration was performed on a single user, considering 625 functional tests distributed across five navigation commands, with 125 attempts per command. The system correctly recognized 591 commands, corresponding to an overall success rate of 94.6 ± 0.9%. The success rate per command was 96.0 ± 1.8% for forward, 88.0 ± 2.9% for backward, 96.0 ± 1.8% for turn right, 94.4 ± 2.1% for turn left, and 98.4 ± 1.1% for stop. The standard deviation of the success rate was estimated from the binomial proportion of correct responses. The overall average functional latency was 176 ± 24 ms. The most stable command was stop, with the highest success rate, lowest latency, and least temporal dispersion (125 ± 15 ms). The command with the highest error rate was backward, with an 88.0 ± 2.9% success rate and a latency of 215 ± 34 ms. These findings suggest that the stability of myoelectric control depended on the reproducibility of voluntary contraction, threshold calibration, and the quality of sEMG signal acquisition.

Conclusions: The proposed method demonstrated preliminary technical feasibility for controlling a robotic system using voluntary muscle activation. Calibration showed a high recognition rate and latency consistent with timings suitable for the intended application. This platform can serve as a scalable foundation for technologies applied to rehabilitation, assistive robotics, and myoelectric control training.