Academic journal article Human Factors

Acquisition of Control Skills with Delayed and Compensated Displays

Academic journal article Human Factors

Acquisition of Control Skills with Delayed and Compensated Displays

Article excerpt

INTRODUCTION

Numerous studies document the effects of delaying feedback for closed-loop, continuous manual control (Ricard, 1994). This feedback has been visual, and the size of the delays studied has ranged from the several-second transmission times of space exploration to the millisecond calculation times of modern computing systems.

For continuous control tasks, human controllers respond to delayed visual feedback by reducing the bandwidth of their input. Such lowering of the frequency, at which the person-plus-system gain curve crosses from greater-than to less-than unity, is called gain crossover regression and has been predicted by modeling (Hess, 1984a) and demonstrated in real vehicles (Berry, Powers, Szalai, and Wilson, 1982), as well as in simulations of them (Hess, 1984b; Riccio, Cress, and Johnson, 1987). The stability of a system is measured by its phase margin - the difference between - 180 deg and output phase at the crossover frequency. Activity in the crossover region (typically 0.5-0.75 Hz) is required for maintaining stability, and delays force a trade of system responsiveness for stability. With very long delays, reducing one's bandwidth is ineffective, and continuous control degenerates to a policy of "move and wait."

Early studies of tracking and remote manipulation showed the problems of control that accompany delays, but it was progress in flight simulation that stimulated efforts to compensate for them. Delay compensation was initially applied to the motions of simulated cockpits. Parrish, Dieudonne, Martin, and Copeland (1973) described an adjustment scheme whereby linear prediction was used to calculate translational positions and a curvilinear scheme was used for the higher-bandwidth rotational positions.

Vigorous flight regimes excite the high-frequency components of a simulated aircraft's response, and wide field-of-view displays make apparent both the delay between stick input and image update and the high-frequency components amplified by prediction. Lead/lag transfer functions have been used to compensate visual delays by balancing the lead and lag terms to band-limit the advance of output phase. The location of the band was empirically set by Ricard, Norman, and Collyer (1976) and Ricard and Harris (1980) and was analytically derived by Crane (1983).

A lead/lag scheme produces small distortions of gain in the crossover region, and subsequent work has tried to reduce these. Baron, Lancraft, and Caglayan (1982) used an optimal control approach to return control with delayed feedback to its nondelayed form, and McFarland (1988) incorporated present position and several past values of velocity to predict position at the end of a delay. This approach takes advantage of high iteration rates, short prediction spans, and the fact that pilot control input rarely exceeds 3 Hz to make accurate band-limited predictions for simulations of flight. Recently, Cardullo and George (1993) compared performance under several prediction schemes using a steady-state, roll-axis tracking task.

Manual control is adaptive, and measures of control input and system performance may appear robust to delays under 100-200 ms. Indeed, vehicles are designed to tolerate a wide variety of inputs and respond with acceptable performance. Yet it is feedback delays of this magnitude for which compensation schemes have been most successful. For these short delays, the development of control skill may be more sensitive to the presence of a delay than is the overall performance of the person-vehicle system.

The present work was done to show that delay compensation benefits not only steady-state performance but also the acquisition of control skill. To that end, delays were added to a compensatory tracking task, and subjects' acquisition times were measured when lead and lead/lag terms were added, after the delay, to adjust the position of a controlled element. In addition, trial averages of subjects' control inputs and output error were used to evaluate performance on each trial. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.