Investigation of kinematics and inverse dynamics algorithm with a DSP implementation of a neural network

An investigation is described to demonstrate the benefits which can be gained by using a digital signal processor (DSP) to implement robot related control schemes, kinematics, and inverse dynamics with a neural network. A neural network adaptive controller is given and applied to a robot manipulator having a closed kinematic chain, a configuration which is not well suited to the popular serial link algorithms. The Lyapunov's stability approach is used to develop a learning rule for the neural network controller that would guarantee the stability of the training process under mild conditions. The controller hardware consists of a PC-386, a fixed point DSP, and a floating point DSP. The software installed on each of these processors has the requirements of satisfying the specific responsibility assigned to that processor and of communicating with other processors so that necessary data is passed on in a timely manner. A computational software package has been built to further enhance the speed of the general control scheme and the neural network algorithm. The techniques used in the DSP implementation of the adaptive control algorithm in real-time are also discussed.