Patient Behavioral Adoption in Self-Correcting Prosthetics via Deep Reinforced Learning: Case of LUROX D

Authors: Taheem Ahmed, Wafi Danesh

SUNY Campus: SUNY New Paltz

Presentation Type: Poster

Location: UU 111

Presentation #: 62

Timeslot: Session C 1:45-2:45 PM

Abstract: The current project introduces LUROX D, an affordable and intelligent self-learning prosthetic arm prototype for amputees suffering from severe neurological disorders that render motor neurons inactive. In essence, LUROX D operates as an end-to-end artificial intelligence (AI) system that combines three sensory phenomena: vision using object detection, speech using speech recognition and thoughts using through recognition, via a deep reinforcement learning (DRL) algorithm to derive intent. By deciphering intent, LUROX D can determine the behavioral pattern of a patient and incorporate it into the prosthetic arm to deliver the correct arm movement action. LUROX D operates as a feedback-based system, where each piece of sensory information: vision, speech and thought, provides the necessary feedback required for learning using DRL. At present, an initial prototype prosthetic arm has been developed that is undergoing extensive training based on both public and private datasets. Preliminary results show a drastic reduction of 70% in error for object, speech and thought recognition. A corresponding increase of 75% has been observed in determining the correct movement pattern of the DRL algorithm. The expectation is that with more specific training data, LUROX D should be able to adapt to the unique behavioral pattern of the patient and forego training completely.