抄録
Human shoulder joint has 3 dgree of freedom, in compact structure compared to general robot arm joint system. This space efficiency is performed by ball joint mechanism in glenohumeral joint. We aim to develop new joint mechanism which has ball joint mechanism driven by wires around rotator. The experimental model was made up on 6 wire model with sufficient moment arm enables to drive it in specified motion area. A Neural network was utilized to learn complex inverse kinematics. However, ball joint system has conjunct rotation known as 'CODMANS PARADOX', it is need for a triaxial joint at the root of a limb. In this paper, we defined spherical surface coordinate system that enables to isolate from conjunct rotation and to be effective learning by Neural Network. This report shows this definition and an experimental result of arm trajectory of 3 dimensional drive of the joint system.
