Learning control of a robot manipulator based on a decentralized position-dependent PID controller

Abstract

The paper describes an approach to learning feedback control of a robot manipulator, based on partitioning of the joint space into segments. Within each segment the robot is controlled as a decoupled linear system by means of conventional PID controllers. To achieve continuity of control variables the segments are represented as fuzzy sets. The controller settings are adapted by online identification from past measurements of position and control signals.