Uma abordagem neuromórfica para o controle tátil do escorregamento em próteses de membros superiores

Abstract

The dexterity of the human hand is a mark in human evolution. Our vast repertoire of manual skills allows us to manipulate objects and tools of different sizes, shapes and weights. This dexterous ability is provided by the biomechanics of the human hand with all the muscles and joints that provide motor function and the sense of touch that incorporates thousands of receptors spread all over the skin. These receptors capture information related to texture, roughness, hardness and shape as well as the location and pressure of contact points over the skin. Unlike their biological counterpart, bionic hands such as the ones used for prosthetics, are not equipped with touch sensors by design. Therefore, manipulating objects is a challenge given that no information can be obtained from the interaction between the prosthetic fingers and the objects. In this scenario, incorrect adjustments of grip force might lead to grasp instability which might cause damage to the objects in case of excessive force or slip if there is insufficient grip force. This thesis presents original contributions to the detection and suppression of object slippage in prosthetic hands. Two distinct methods are presented where both employed neuromorphic principles in the development of artificial tactile sensing systems. Tactile information is coded in spikes that resemble the activity of mechanoreceptors and are used as feedback signal to a control system that behaves as a reflex that assists in the control of grip force. Both methods were inspired in the activity of FA-I afferents that encode transient tactile stimuli, displacement on the skin and slip. The first method makes use of an optical sensor that converts object displacement into spikes that are used as input to a monotonic PI controller. The second method makes use of a multitaxel tactile sensor in a 4x4 matrix format. Force signals are converted into spikes using the Izhikevich model to mimic the activity of FA-I afferents. First, it was demonstrated that spikes generated continuously during motion of the object could be used as input to a controller that effectively suppressed object slippage. The second method was developed based on the observation that the first spikes triggered by the tactile afferents carry enough information to trigger motor responses. Therefore, it was possible to develop a bioinspired method that detects slip based on transient events such as generated by force signals. This approach does not make use of continuous spikes, prompting advancements with respect to the previous method. This thesis presented two diferente methods that can be used to develop novel promising sensory capabilities to prosthetic hands. Incorporating reflex actions will enhance the usability of such devices, increasing their acceptance so that users can use the bionic hands with more freedom and confidence when performing different daily tasks.