Representation of the human operator as a sampled-data system

摘要

An experimental and theoretical investigation is described which led to the formulation of a sampled-data model that accurately simulates the characteristics of the human operator in a compensatory tracking task. The term ?compensatory? applies to a restricted situation in which the operator could directly sense only the difference between his actual output and that defined by a reference input. The task actually studied was further restricted to visual sensing of error and manual output by horizontal movement of a vertical control lever. Error was displayed to the operator as a proportional horizontal displacement of a vertical line on a c.r.t., and his task was to keep this line as close as possible to a central zero-reference point; i.e. to track the input with the smallest possible error. Study of the effects of interposing a sample/hold operation prior to the error display indicated that the operator acted upon samples of position and velocity error taken at intervals of approximately 0.15s. The nature of the operators' responses to step and ramp inputs was in accord with this hypothesis and led to the initial formulation of a sampled-data model. Comparison of the responses of operators with those of the model in a continuous tracking task revealed some deficiencies and the model was elaborated by incorporating feedbacks to simulate more sophisticated prediction. It then proved capable of very accurate simulation, so that it was possible to study the operators' open-loop tracking behaviour by using a synthetic error signal derived from the model. The results confirmed that the basic structure of the model was correct. Some discrepancies were also apparent: these were mainly explained by the minimisation of complexity consistent with reasonably accurate simulation. It was concluded that, in the task studied, the behaviour of the human operator could be characterised by a random sampling of both position and velocity, followed by a computation of required futur-ne hand movement (according to some optimal strategy) during the succeeding sampling interval and its subsequent execution over the next sampling interval. Each sampling action seems to take from 20 to 45ms and is followed by an interval of 80?220ms before a subsequent sample can be taken.