Estimation of prosthetic arm motions using stump arm kinematics

This paper proposes two kinematic based task classification methods to aid control of a transhumeral prosthesis. The first method is a neural network based classifier where the angles of shoulder flexion/extension, shoulder abduction/adduction and elbow flexion/extension are considered. The angular values with their first and second derivatives are obtained to train the robotic arm for a selected set of tasks. The second method uses a fuzzy logic based classifier where the angles of the shoulder and elbow motions are divided into angular positions such that each combination of the above motions performs a specific task. Therefore, more tasks can be defined with the combinations of the angular positions of the motions. The effectiveness of two task classification methods is verified experimentally.