Vehicle control may be considered as a hierarchically structured set of functions. Plan conception and plan selection activities are performed in the navigation function, verification and adjustment of the short-term voyage progress are performed in the guidance function, and typical closed-loop control activities are performed in the control function. Supervisory control of vehicles deals with automated vehicle control functions to a large extent. Nowadays, technology permits the typical control functions to be entirely executed by automated systems, while the navigation and guidance functions are still partially automated. The operator, who may observe the controlled process directly, acts as a manager who supervises the system and only interacts with the automated system by performing corrective actions. The human operator remains the primary responsible factor, specifying the constraints, and procedures in terms of setpoint changes for the automated system. It is known, however, that humans have certain limitations in supervising capabilities, particularly when slowly responding systems are concerned. Humans are not always able to anticipate, that is, to mentally predict future state information by applying knowledge of the goals, process characteristics and disturbances that act on the system, considering the current control actions and the observed changes in process state. Literature on human performance models indicates that accurate vehicle control mainly depends on the operator's knowledge of the controlled process characteristics and the route to be followed, on the ability to predict the future states of the controlled process, and on the quality of track planning and track following activities. A technical system to support operator anticipating behaviour should provide information that addresses these elements. The current study focusses on the question what type of information is required to obtain accurate supervisory vehicle control. Five experiments were conducted in a man-in-the-loop simulator facility, where participating subjects were required to perform specific aircraft and ship control tasks, in routine and in non-routine task conditions, using different information systems. Task performance was measured and analysed with respect to accuracy of guidance and quality of navigation. The need of three information elements for accurate supervisory control was investigated: (i) preview information that enables immediate perception of the planned route, and assessment of future position error relative to that route; (ii) path prediction information, showing the predicted track ahead; and (iii) capability prediction information, showing the predicted consequences of a range of process control actions with respect to the process states reaching their predetermined safety margins, thus indicating the boundaries of safe operation. The results of the experiments indicate that accurate supervisory vehicle control is obtained when information is provided containing the following complementary elements: (a) Capability prediction information, needed to support the navigator in the navigation function (knowledge-based behaviour), where non-routine task conditions occur. Capability prediction aids the mental process of generating alternative solutions for a problem situation. These alternatives are presented as an advise to the navigator. Knowledge of the safety margins allows navigators to properly re-plan the route by selecting the safest alternative. (b) Preview and path prediction information, needed to support accurate guidance along that alternative route (rule-based behaviour). During guidance, only routine task conditions exist. Navigators first roughly pre-set their controls based on preview and rules-of-thumb for control, then perform corrective setpoint control actions based on error information provided by the path predictor. It is suggested to apply the findings of this study for the guidance and navigation of highspeed craft. Because continuous observation of the outside world and of the controlled process state is required, it is advised to show prediction information as an overlay, superimposed over the external world scene, thus indicating where safe areas of navigation are located. The present study focussed on anticipation behaviour in guidance and control functions with a limited time horizon, for example, 5 to 10 times the time constant of the controlled process. Many navigation functions exist that encompass larger time horizons. In that case, more factors affect the controlled variables, requiring more knowledge based support. Additional research is needed to assess these longer term support systems more objectively. It is recommended to improve technology for prediction. For example, an extrapolator is relatively simple to implement because extrapolators operate on the basis of a limited quantity of sampled data. In practice, however, it is questionable whether sensors are capable to provide signals with sufficient stability and accuracy. (A)
Abstract