Developing Prosthetics with Deep Learning and Soft Robotics
Disciplines
Artificial Intelligence and Robotics | Computer-Aided Engineering and Design
Abstract (300 words maximum)
Recent advancements in the fields of soft robotics and artificial intelligence have allowed for the use of novel techniques in addressing the challenges experienced by the more than two million amputees in the United States. Such challenges include controlling myoelectric prostheses, which require an extensive amount of training and calibration to use. Such designs result in modern prosthetics being incredibly expensive, along with increasing the risk of overuse injury and device rejection by the amputee. Therefore, we propose a novel design which uses machine vision and soft robotics to create an automatically grasping prosthetic. This prosthetic uses a convolutional neural network for identifying the appropriate grasp type for the object in question, which will then call a second neural network which is trained on finite element data to finely control the movement of the prosthetic. The prosthetic itself is designed with soft robotics in mind so that it grasps and functions more similarly to that of a human hand. The final objective of this project is to create a fully functional prosthetic which aims to reduce the overall costs of prosthetics by serving as a basis for which future human prosthetics can be built off of and commercialized. This will in the long run, increase the quality of life for people with amputations by providing them with more adaptable and affordable prosthetics.
Academic department under which the project should be listed
SPCEET - Engineering Technology
Primary Investigator (PI) Name
Turaj Ashuri
Developing Prosthetics with Deep Learning and Soft Robotics
Recent advancements in the fields of soft robotics and artificial intelligence have allowed for the use of novel techniques in addressing the challenges experienced by the more than two million amputees in the United States. Such challenges include controlling myoelectric prostheses, which require an extensive amount of training and calibration to use. Such designs result in modern prosthetics being incredibly expensive, along with increasing the risk of overuse injury and device rejection by the amputee. Therefore, we propose a novel design which uses machine vision and soft robotics to create an automatically grasping prosthetic. This prosthetic uses a convolutional neural network for identifying the appropriate grasp type for the object in question, which will then call a second neural network which is trained on finite element data to finely control the movement of the prosthetic. The prosthetic itself is designed with soft robotics in mind so that it grasps and functions more similarly to that of a human hand. The final objective of this project is to create a fully functional prosthetic which aims to reduce the overall costs of prosthetics by serving as a basis for which future human prosthetics can be built off of and commercialized. This will in the long run, increase the quality of life for people with amputations by providing them with more adaptable and affordable prosthetics.