Dynamic control of manipulators using qualitative and quantitative modelling

In this paper we present an application of fuzzy techniques for the dynamic control of manipulators. The system proposed is divided into two levels. The low level is shared between classical methods of control and a fuzzy scheduling approach. The high level is a task dependent level where fuzzy controllers are used in a modular way. The low level is designed as follows: We start from an identified quantitative model of a robot. We consider the control space defined by the control inputs limits. In this control space we choose “points” depending on the desired task. For each point, the dynamic model is automatically linearised. For each linear model obtained we build a Proportional, Integral, Derivative controller using a linear quadratic method. For each point the value of the PID gains and the torques obtained from the non-linear model are stored. During control a fuzzy scheduler is used to compute the values of both torques and gains and to send them to the robot. For the high level, fuzzy controllers inputs of which may be joint accelerations, velocities and position torques or payload variations are designed depending on the task. The qualitative rules used for these fuzzy controllers are obtained from task constraints. This supervision system allows us to reduce tracking error where or when it is needed.