The SACE process runs in parallel to and complements the activities undertaken as part of an existing systems engineering and associated system safety assurance process. We assume that such a system safety assurance process is in place and that the activities described as part of the SACE process are modifications, enhancements or additions to that system safety process to specifically deal with the safety assurance challenges of an autonomous system operating in a complex environment. In this guidance we do not describe what that baseline system safety assurance process is, however, neither do we assume that any particular system safety process is adopted, instead we define certain characteristics that the safety assurance process should have. Figure 1 below provides a model of this baseline safety process that illustrates the following required features:
These features are influenced by established good practice system safety processes such as the Aerospace Recommended Practice ARP 4754A [2].
Note that the system safety assurance process itself runs in parallel to, is informed by, and informs, the system development process. In addition, the development of the safety case for the system also runs in parallel to this process. The safety case process is discussed in more detail below. For clarity we do not show either the development process or the safety case development process in Figure 1.
This SACE guidance covers only part of this overall system safety assurance process, as indicated by the SACE box in Figure 1. SACE starts at the beginning of the development process when a concept for the AS has been determined and continues down to the derivation of requirements for the sub‐systems of the AS. SACE also considers verification of the AS at the sub‐system and system level. SACE does not include the development of requirements for the individual system components, or the implementation of those components. These issues are considered as part of other guidance documents. For example, safety assurance of components implemented using machine learning is considered as part of the AMLAS (Assurance of Machine Learning for Autonomous Systems) guidance [20], as indicated in Figure 1 below.
There are other safety assurance aspects that are not within the scope of the SACE guidance. Firstly, the safety assurance activities considered by SACE are only those that are applied during the development phase of the AS lifecycle, that is activities carried out prior to deployment of the AS into operation. Although the SACE activities will have consideration for safety assurance during later lifecycle phases such as operation and maintenance, the safety assurance activities that are actually undertaken during these later phases (often referred to as operational safety management) are not within the scope of this document. Secondly, SACE focuses on safety assurance for an individual AS. Although this involves consideration of the interaction of that AS with other agents (including other AS), SACE does not explicitly consider the additional safety assurance implications of multiple collaborating ASs (see [14] for examples of safety approaches for collaborative robots). Thirdly, SACE does not provide guidance on the legal and ethical considerations surrounding the development and operation of AS. Such issues will be considered as part of separate guidance [38]. SACE takes as an input the output from such considerations in the form of defined acceptance criteria. This input to the SACE process is discussed when describing Stage 2.