Roles, Responsibilities and Task Management

Crew task management (CTM), including proper prioritization and allocation of tasks between crewmembers, is another critical factor in flight safety. In a review of NTSB and ASRS reports [31], task management errors occurred in 23% of the accidents (79/324) and 49% of the incidents (231/470) reviewed. Proper task management strategies address not only who is or should be performing a task, but that crewmember’s progress on task, how many tasks are in his queue, the complexity of the task (and whether it is possible for the human or IA to “multitask”). These factors will determine the need for task switching or reallocation at any given time.

Task management is especially critical to human-IA interaction, because there is a wider disconnect between the way information is ingested, evaluated and used to support decision-making between a human and a machine. To understand this, it is helpful to look at several ASRS case studies of aircraft incidents which involved task management errors.

Case Study 1

In the first case, a DHC-8-400 (Q400) crew distracted by weather, turbulence, EFB usage and pas- senger safety, forgot to reset a low altimeter (29.11 to 29.92) while climbing to FL230 and were subsequently advised by ATC of an altitude deviation.

The aircraft had been delayed for maintenance before departure and had the passen- gers on board for an hour, so there was a temporal demand factor. There were multiple lines of thunderstorms and high winds present on departure, distracting the PF. The PM was having some problems with a frequency change (the reception was poor and they were given a wrong frequency) which took some time to resolve. Both pilots were dis- tracted and failed to reset either altimeter, which resulted in a large indication error. As they checked on to the new frequency, the Controller asked what their altitude was, which is when they spotted and corrected the problem. The main factor in this event was attention tunneling; the PF was focusing on the EFB to find clear weather and the PM focused on frequency change. Four task management errors observed in this case included (1) low prioritization of the instrument monitoring task (failure to follow SOP on departure) which should have been a priority (“aviate”) task, (2) excessive monitor- ing of the weather (PF) and radio (PM), (3) low resource allocation to the aviate and monitoring tasks, and (4) high workload. Contributing factors included the temporal de- mands imposed by the ground delay, severe weather, and stress (expressed by the PF in his report).

Case Study 2

In this case, B737-700 flight crew experienced difficulties attempting to practice a RNAV approach into JAX in Visual Meteorological Conditions (VMC). Control Wheel Steering (CWS) mode was inadvertently selected and the aircraft descended prematurely triggering a low altitude alert from 109

7 V&V ISSUES AND APPROACHES

ATC and an EGPWS warning. A visual approach ensued.

It was the Captain’s (PF) first attempt at practicing an RNAV approach and the first officer (PM) had only flown one before. The crew was on approach into JAX, and re- quested the RNAV (GPS) 31. ATC cleared them direct to NIBLE (IAF); however they were past NIBLE when starting the turn which resulted in a larger turn than expected (virtually entering a downwind). When cleared for the RNAV approach, due to the close proximity to NIBLE, the autopilot was turning right towards the fix and shortly thereafter started a left turn towards POTME (close timing was a factor). At some point, the PF unknowingly disengaged LNAV and went into CWS. PM alerted him to the mode change, but he did not respond. The crew became disoriented; they were off course, confused with what the airplane was doing, and descending. The PF disengaged the autopilot and continued descending while flying away from the field. PM instructed him several times to stop descending, climb, and return to the glidepath. The PF, fixated with the new RNAV procedures, was unresponsive to the PM’s verbal alerts. The AC had descended below the charted altitude at POTME (2,600 ft, 9 miles from TDZ) to about 1,700 ft MSL. ATC alerted the crew, who responded with “correcting,” requested and were subse- quently cleared for the visual approach. The main factors in this event were poor CRM, resource mismanagement and unfamiliarity with the RNAV procedures. The PM noted in his report that they should have prioritized flying the airplane above all else, and should have discontinued the RNAV approach earlier when they became disoriented. The PM felt he was assertive but nonetheless, the PF failed to heed his guidance. His fixation on the procedure and mode confusion caused him to ignore (or perhaps fail to hear) the warnings of his crewmate.

Four task management errors we observed; these include (1) high prioritization of the RNAV procedure, (2) excessive task monitoring (procedural as well as confusion about the AC control response), (3) failure to initiate the climb and redirect to course, and (4) failure to terminate the RNAV procedure and request the familiar VFR approach as soon as they became disoriented. Other contributing factors included inadequate training, poor CRM and spatial disorientation.

Case Study 3

In the third example, an A320 Captain on the LAS TYSSN THREE RNAV calculated his descent constraint compliance incorrectly and failed to meet one of the constraints before realizing his error, even though his First Officer was cautioning about the error.

Enroute to LAS, ATC cleared the crew to descend to FL260 at 280 KIAs. The crew leveled off at FL260 and set cruise mode for a managed descent speed of 282 KIAS; however, the PF erroneously selected 280 on the FCU instead of 280 KIAs. ATC cleared them to descend via the RNAV arrival for a visual approach to Runway 25L. The PF dialed in 8,000 to meet the altitude restriction at PRINO. Both crewmembers had triple 110

7.5 Challenges and Approaches for Simulation of Human-IA Interactions

checked all the altitude constraints on the arrival, and both had the constraints button pushed. The first waypoint had an “at or above FL200” constraint and the subsequent waypoint had an “at or below FL190” constraint. They engaged managed descent and calculated that they needed 21 miles to get below FL190 over the second waypoint (19 nm away from the first). The aircraft started a 900 FPM descent, still in selected speed of 280. The MCDU showed approximately 2,400 FT low on the path but slowly decreasing deviation. The PM noticed and informed the PF about the navigation errors, but due to attention tunneling (focusing on the altitude and descent rate calculations), he failed to adhere to proper procedure to correct the descent.

Three main task management errors were observed, including (1) low prioritization of the aviate task, (2) late initiation of the corrective measure, and (3) low/poor resource allocation in not prescribing the troubleshooting to the PM. Contributing factors included distraction, over thinking on the part of the PF, and poor CRM overall.

The three cases presented above illustrate the impact of task management errors on the pilots’ abil- ity to properly navigate, adhere to published procedures and maintain stability on the flight path. Procedural compliance and FPM are directly impacted by crew task management errors.

Another important consideration in the assessment of a cooperative human-IA crew is that of role- setting. The relationship between a crew of two human pilots in terms of responsibilities and roles may vary from day-to-day and even during a single flight. There are four cornerstone concepts relating to the design of cooperative man-machine systems [52]. These are: ability, authority, control and responsibility (Figure 23). Consistency in the relations between these concepts has been identified as an important quality for the system design.

Figure 23: Authority and Autonomy in Cooperative Human-IA interaction

But IA systems designed to work interchangeably with the human crewmate must be more adaptable to changing situations. Simulations should test to ensure that the requirements expressed in the figure above are met for both the human pilot and the IA system, across all operational scenarios. To this end, simulations designed to validate the IA must not only ensure that the highest level of responsibility can be assumed by the IA, but that the highest level of responsibility may also be assumed out by the human pilot, and the IA system will react in an appropriately dominant or passive manner.

7 V&V ISSUES AND APPROACHES