Fuzzy adaptive controller design for the joint space control of an agricultural robot

The proper execution of agricultural robotic tasks needs the use of adaptive control techniques. This fact is mainly due to the nature of the systems to control, which are difficult-to-model and time-varying systems. After a review of previous works concerning adaptive control, a solution using a fuzzy adaptive controller is studied for the joint control of such robots. An analytic representation of a particular fuzzy system is first developed to deduce useful conclusions for the controller design. Then, a specialized learning architecture is used to allow the reconstruction of an error signal required for a gradient method for on-line modification of the consequent part of the inference rules of a Sugeno's fuzzy controller. At the same time, a second level constituted by static rules (meta-rules) is introduced to cope with some limits of the learning architecture. Clustering of some rules is proposed to be able to learn those that are not fired most of the time but essential for unusual robot motions. Thanks to this new structure, the controller is dedicated to each meta-rule, and the number of rules with respect to a solution without meta-rule is considerably reduced. Simulation results during large on-line variations in system parameters derived from a typical example of an agricultural robot show the effectiveness of the proposed approach. The controller stability is verified by using the so-called cell-to-cell mapping algorithm. Finally, the feasibility of the implementation of this algorithm in low-end hardware is shown. © 1998 Elsevier Science B.V. All rights reserved