A Low-Cost Arm Robotic Platform based on Myoelectric Control for Rehabilitation Engineering

Rehabilitation robotics is a recent kind of service robot that include devices such as robotic prosthesis and exoskeletons. These devices could help motor disabled people to rehabilitate their motor functions, and could provide functional compensation to accomplish motor activities. In order to control robotic prosthesis and exoskeletons it is required to identify human movement intention, to be converted into commands for the device. Motor impaired people may use surface electromyography (sEMG) signals to control these devices, taking into account that sEMG signals directly reflects the human motion intention. Myoelectric control is an advanced technique related with the detection, processing, classification, and application of sEMG signals to control human-assisting robots or rehabilitation devices. Despite recent advances with myoelectric control algorithms, currently there is still an important need to develop suitable methods involving usability, for controlling prosthesis and exoskeletons in a natural way. Traditionally, acquiring EMG signals and developing myoelectric control algorithms require expensive hardware. With the advent of low-cost technologies (i.e. sensors, actuators, controllers) and hardware support of simulation software packages as Matlab, affordable research tools could be used to develop novel myoelectric control algorithms. This work describes the implementation and validation of a Matlab-based robotic arm using low-cost technologies such as Arduino commanded using myoelectric control. The platform permits implementation of a variety of EMG-based algorithms. It was carried out a set of experiments aimed to evaluate the platform, through an application of pattern recognition based myoelectric control to identify and execute seven movements of the robotic upper limb: 1-forearm pronation; 2- forearm supination; 3-wrist flexion; 4-wrist extension; 5- elbow flexion; 6- elbow extension; 7-resting. The algorithm use a feature extraction stage based on a combination of time and frequency domain features (mean absolute value, waveform length, root mean square) and a widely used k-NN classifier. Obtained mean classification errors were 5.9%. As future work, additional features in the myoelectric control algorithm will be evaluated, for real-time applications.