No abstract

No abstract

No abstract

Different approaches of a learnable Fuzzy-Logic (E) concept for highly fast and accurate position control of industrial robots will be presented which provides both time-optimality for large position errors as well as well damped response (no overshoot) near the target. For automatic optimization of the control parameters an additional Neural-Network component is introduced. Based on simulation and cxpcrimcnts the performance and robustness of the presented FL concept are discussed.Within the last 15 years many powerhd robot control concepts and algorithms have been proposed which, without exceptions, are based on analytical robot models. Based on these mathematical models servo controllers can be designed which can base on different optimization criteria. However, the practical efficiency of these modelbased control concepts depends strongly on the accuracy in which the model represents the real static and dynamic system behavior [ 11.Due to the strong kinematic nonlinearities, to the dynamic coupling of adjacent robot joints as well as to the unavoidable elasticities of the mechanical structure in some cases robot models are very complex and clumsy. Thus model identification by means of experimental system analysis can be very expensive or can't be realized in the worst case. Moreover, almost all methods of model identification are based on the simplified assumption of linear system response. Thus control concepts based on such models obviously have only a limited area of validity.In such cases control concepts based on analytical models can provide insufficient cost/performance ratio, a poor robustness with respect to model uncertainties or variation as well as a lack of physically clear transparency of their functionality.Thus there is an increasing demand for alternative control concepts which do not presume an analytical robot model, are based on unsharp ("fuzzy") linguistic description of the measured process state and control input signals, can be specified and programmed by a set of heuristic application oriented rules and have the optimal capability of learning or adapting their control parameters to the desired optimum. Rule-based Fuzzy-Logic (FL-) IEEE International Conference o n Robotics a n d Autoniatlon -1184 0-7803-1965-6/95 $4.00 01995 IEEE controllers seem to comply with these steadily increasing requests.Within this paper several structurally different FLconcepts of robot position control as well as Neural Network (NN) component for control parameter optimization will be introduced. Their performance and robustness will be discussed by simulation and experimental results. Control problemA very important objective of point-to-point (PTP-) movements of robots is the ambiguous demand -to run as fast as possible in the case of far target distances (time optimality) and-to approach the target smoothly without overshoot collision avoidance).As illustrated by the simulated step responses of a robot joint in Figure 1.1 a linear servo-control concept cannot achieve this desired system behavior. Either ...

No abstract