An architecture for a motor control system inspired by biological organisms is outlined. The core of this architecture is a model of the direct kinematics of the articulated chain (AC) under control. The advantage of using the direct kinematics solution to solve the inverse kinematics problem is that the former is separable and can be broken down to low-dimensional problems. A novel algorithm to adaptively learn, in a hierarchical fashion, the direct kinematics solutions of an AC with many degrees of freedom (DoF) is presented. The algorithm is designed such that only neurally implementable operations or functions are used. The algorithm is shown to work with an articulated chain with nine DoF. On average, less than 200 iterations per joint are required.
展开▼
