Abstract
Robots are revolutionising the field of prosthetics. With the application of machine learning, prosthetic rehabilitation can become more natural in terms of performance and sustainable for human anatomy. In this study, the human activity of scooping various forms of materials is investigated.
The activity is divided in different motion primitives. For each primitive, human right arm degree of freedom to jaco human skill is analysed in terms of physiological parameters, including wrist pronation-supination angle, elbow flexion angle, shoulder rotation/abduction/flexion angles, hand acceleration, and wrist pronation angular velocity.
Optimum human posture is identified for each primitive in terms of muscular effort using musculoskeletal modelling. This analysis identifies how humans execute same activity for the different types of materials. Each human motion primitive is mapped to robotic arm primitive for efficient execution for different materials. The 6 DOF robotic arm is equipped with a camera to capture the image of the material which it has to scoop. The robot is programmed using a deep convolutional neural network, which is trained to identify the type of material using machine vision. Consequently, the activity can be performed efficiently based on human intuition in a dynamic environment. Hence the proposed research fully utilises the mechanical potential of robots within the constraints of human musculoskeletal system.