Response to VSST2-1: transition from automated to manual flight: avoiding the surprise "automation surprise" represents phenomena in which the behavior of the automation diverges from the intentions…

Response to VSST2-1: transition from automated to manual flight: avoiding the surprise "automation surprise" represents phenomena in which the behavior of the automation diverges from the intentions of the operator. Most automation surprises occur within the time and space margins of the operational envelope and away from obstacles. These result in increased workload and decreased efficiency. When automation surprises occur within hazardous space-time margins or near obstacles, automation surprises directly affect safety and are of serious concern to the aviation community. By definition loss of control accidents due to aerodynamic stall reflect scenarios in which the intent of the operator (to remain within the operational envelope) is not supported by the automation and reflect a dangerous class of automation surprises. This includes automated flight as well as the transition from automated to manual flight. A detailed analysis of the interaction between the operator and the automation in loss of control accidents due to aerodynamic stall in which crew error was cited as a probable cause, identified several different classes of automation surprises including the impact of sensor failures, mode confusion, and command mismatches. This analysis highlights that one-size-fits-all approaches to detecting and mitigating automation surprises cannot work. This proposal describes research to extend the analysis of loss of control accidents to categorize the classes of automation surprises. Based on characteristics of each category, solutions to detect and mitigate this phenomenon can be developed. Two examples: 1) automaton surprises that result from AFM limitations or manufacturer bulletins can be detected by monitoring sensors and automation states and a data-base of procedure limitations. 2) automation surprises that result from mode confusion can be detected by evidence-based hypothesis testing (i.e. probabilistic checking) of the likelihood of correct mode given the aircraft and automation states. The proposed solutions extend state-of-the-art pilot associates and vehicle health monitoring systems by linking the checking of the automation to mission related limitations on procedures and by application of probabilistic approaches to performing crosschecking between distributed automation states. We propose to develop a prototype aircraft pilot associate device (APAD) designed to improve automation awareness by providing pilots with continuous monitoring and alerts, context dependent guidance, and just-in-time training. This program could operate as an integrated part of an electronic checklist or electronic flight bag or as a stand-alone application on a mobile computing device (e.g., ipad). Automation surprise may result from pilots having an inadequate or mistaken mental model of the machine S behavior. Preliminary work suggests that current automation training does not provide pilots with the mental framework needed to build the necessary mental model to understand the automation system. If the lack of accurate mental models of the automation is a primary cause of automation surprise, then the likelihood of automation surprise can be decreased through training that provides accurate functional models of the automation. We propose to develop and test a prototype automation education program to be incorporated with the APAD to guide pilots in creating appropriate mental models of their automation systems. The benefits of this proposal are the explicit address of real-world documented problems and analysis of nextgen scenarios through a detailed analysis of the functional causes, yielding nuanced solutions to each different class of automation surprises.