【正文】 ceeds 5 . For larger deviations, thesystem turned out to be too slow to pensate. The outputvalues are therefore amplified to give the maximum outputcurrent already in the case when the controller wants to applya positive or negative MEDIUM control force. Although weimplemented a matrix with 49 rules, we effectively useonly the core matrix. This leads to another deteriorationof the controller performance.VII. SUMMARY AND CONCLUSIONThe proposed vision feedback fuzzylogic controller is ableto keep the inverted pendulum in the upright position, thoughfor a limited time. In what follows, we will discuss modificationsthat might lead to an improvement in the behavior ofthe experimental system.The fuzzylogic controller implementation is entirely writtenin Borland C . This programming language running onMSDOS operating system does not provide a realtime environmentfor the experiment and results in reentry problemsof subroutines. This problem could be avoided by introducingrealtime subroutines that are not part of the standard Cpackage. To control the inverted pendulum successfully forlong periods of time, we would have to make use of not onlythe angle of the pendulum, but also the position and perhapsthe velocity of the sled. Using the position of the sled also willin turn lead to a fourdimensional fuzzylogic control system.The addition of an extra state, however, will render the manualtuning approach to determine the fuzzy set values a morechallenging task. It might even be necessary to implementan adaptive fuzzylogic controller that obtains suitable controlparameters after a period of “training” [11], [12]. This willrequire major changes not only in the controller program, butalso in the data acquisition system, since the sled position isnot part of the present control strategy. Further problems arisefrom the fact that we use the difference between two anglesas input for the angular velocity. Since the measurementsare noisy and have a relatively large quantization error, weend up puting an even noisier estimate of the velocity.Introduction of filtering techniques in the vision system canpotentially result in a better system performance because of asmaller angular velocity error.The mechanical time constants of our experimental systemlimit the response time and prevent fast changes in themovement of the sled [1], [3], [7]. To achieve an improvement,it would be beneficial to change the design from the leadscrew – table to a beltdriven configuration that is moreresponsive, has a smaller mass and less friction. Another aspectis the use of a different camera to accelerate the acquisition ofdata through the vision system. The Intelledex Vision Systemis able to give a reset pulse to the video camera. This wouldmake it possible to acquire more than just 60 frames per secondand also decrease the response time of the system. Also weencounter problems that result from the limited resolution ofthe video camera. The fact that we have only 480 pixels forthe entire horizontal field of view leads to large quantizationerrors and a low precision. It would be extremely beneficialto improve the accuracy of the measured angle, since theprecision of the input data is of great importance for thecontroller.After reviewing the experimental results, it seems clearthat it would be necessary to make substantial changes toachieve a more satisfactory performance of the vision feedbackfuzzylogic control system. For example, better inferencerules (perhaps adaptive) in the fuzzylogic controller that useposition and velocity information of the sled, along with afaster and more precise vision data acquisition system wouldcertainly lengthen the time that the pendulum is kept in theupright position.