Brain-machine interface for actuating a bionic prosthetic arm

Recent progress in neurobiology go hand to hand with the development of artificial intelligence. The latter in fact, allows to decode the complex patterns underlying the brain activity recorded by new generation implanted electrodes. This combination represent the turning point in the development on neuroprosthetic. In the past decades we observed huge advancements in this field, with paralytics patients driving wheelchairs by visual stimulation, monkey controlling robotic arms, or patients moving a cursor with their minds to write on a virtual keyboard. However no project ever tried to be as ambitious as some are doing be now, such as Neuralink or B-Cratos, that aim at pushing to a new limit the concept of neural interface. The new challenge for the next decade then, is to develop closed-loop decoders, that not only could allow fine controlling of bionic prosthesis exploiting intra-cranial EEG recordings (with their low signal to noise ratio), but also to provide sensory feedback as response, replicating the realistic neural response in the patient brain. In this thesis I am investigating the possibility of controlling a 3D printed prototype prosthetic arm, to replicate the movement performed by a macaque monkey during a grasping trial. The dataset used is composed by the spike-sorted brain activity of the animal, recorded using implanted electrodes arrays, from researchers of German Primate Centre. Many networks and configurations were tested, before assessing that the recurrent ones were those performing the better. In particular was successful the idea to allow a convolutional layer to extract spatial features from unprocessed data and have an LSTM weighting temporal dependencies. Such a network was then used to build the brain computer interface, used for a proof of concept of low delay online decoder. In the demonstration, the state of the experiment, the object shape and size, are decoded, translated to a proper actuation binary radio-command and sent the arm.