An adaptive neural controller for a robot

摘要

This article describes the implementation of a neural net based trajectory following control for a simplified model of the Puma 560 manipulator robot, assuming the usage of electric drivers and taking into account the flexibility in the transmissions. The control strategy uses a PD controller with an additional feed-forward neural net portion, which is responsible for obtaining the robot's inverse dynamics and is trained using a learning algorithm with adaptation in real time.