Situation Awareness: The Key to Safe Autonomous Systems: Plenary Talk

Abstract

Human trust and Situation Awareness (SA) are safety critical components of highly automated systems. May it be an autonomous surgical robot or a self-driving car, independent decision making of the system based on complex, multi-sensory data will surly lead to some wrong conclusions and hazardous outcome. The aim of the development community is to establish processes and metrics for ensuring the safety of such systems. Safety begins at the development level, where new guidelines and requirements are being established to support manufacturers in transparency of development and production, leading to better accountability. New models define the Levels of Autonomy for such systems, linking the engineering functions to hazards and essential performance criteria [1]. Due to the nature of the novel functionalities, users are required to understand the limitations of these systems, and the role of the humans driver can be critical in such systems. Soon enough, fully automated vehicles (SAE Level 5) will take the complete responsibility off of the driver, but partially and highly automated systems (L3-4) may require human intervention in emergency situations or in driving scenarios outside the operation domain (ODD). The transition between automated and manual driving is an understudied, yet safety critical process. When the driver has to take back control from the automated system, a handover situation occurs. The request for initiation is carried out at the time instant takeover, and the time between the request signal and the actual retake of human control is called the takeover time [2]. The efficiency of hand over processes largely depends on the environmental conditions and the human's perception capabilities, modeled by SA, i.e., the dynamic understanding of the scene and possible actions to take. SA can be assessed along two major criteria, based on the level of understanding and affecting components. In the past years, an incremental increase of the level of autonomy could be observed in the automotive industry. As a tendency, by adding new functionalities to existing features, autonomous capabilities shift from L2 ADAS systems towards L3 partially automated and L4 highly automated systems. Given the incremental advancement of L2-3 features, users tend to underestimate the need for SA, as the system seemingly takes over the supervision of certain components. Based on the state of the environment, three levels of SA are defined [3]. At Level 1, the driver is responsible for perceiving the elements in the environment relevant to the dynamic driving task, i.e., distinguishing between important and neutral actors and objects. At Level 2, there is a need of comprehension and conceptual understanding of these elements relative to the driving task. This includes the assignment of behavioral models to the actors, and their current dynamic states. At Level 3, the driver is required to project the state of the environment after a particular action, i.e., predict actor states, including the dynamic behavior of the ego-vehicle. SA can be decomposed into key components, and full SA means regaining all three SA levels and all five components: • Spatial awareness: knowledge of object locations; • Identity awareness: knowledge of salient items; • Temporal awareness: knowledge of the dynamic states; • Goal awareness: knowledge of the maneuvering plan; • System awareness: knowledge of the environment. Situation Awareness plays a key role during handover processes, as it defines the level of cognitive understanding and capability of a human operator in a given environment. Assessing, maintaining and regaining efficiently SA are core elements of the relevant research projects, reviewed and compared in this talk.