An AI -Based Model for Texture Classification from Vibrational Feedback: Towards Development of Self-Adapting Sensory Robotic Prosthesis

This paper presents a novel method of tuning vibration parameters to elicit specific perceptions of texture using vibration artefacts detected in EMG signals. Though often used for prosthetic control, sensory feedback modalities like vibration can be used to convey proprioceptive or sensory information. Literature has shown that the presence of sensory feedback in prosthesis can improve embodiment and control of prosthetic devices. However, it is not widely adopted in daily prosthesis use, due in large part to the daily change in perception and interpretation of the sensory modality. This results in daily parameter adjustments so that sensory perception can be maintained over time. A method therefore needs to be established to maintain perception generated by modalities like vibrations. This paper investigates modulating the vibration parameters based on how the vibrations dissipate in the surrounding tissue from the stimuli. This is with the aim of correlating dissipation of vibration to specific perceptions of texture. Participants were asked to control vibration motor parameters to elicit the perception of three different grades of sandpaper, provided to them for reference. Once the vibration parameters were chosen a CNN algorithm identified and categorized the artefact features along equidistantly spaced EMG electrodes. Participants were asked to repeat this experiment on three separate days and on the fourth was asked to complete a texture identification task. The task involved identifying the texture of the sandpaper based on their previously chosen parameters and compared the results to tuning against an AI-based algorithm using the dissipation of the vibration artefacts.