A neural network system that controls and plans paths for a robot

Proposes to solve the problems of direct/inverse kinematics and control of trajectories by multilevel perceptrons. The authors´ solution admits a parallel implementation in real time. It does not need either to solve kinematic equations or robot trajectories, because it learns gradually by examples adaptively. The control system consists of different networks each of which specialises in solving a particular problem. This structure enables a modular approach to the problem accelerating convergence. The system obtains an acceptable trajectory and gives a parallel solution that could be used in real-time applications