Defining an Effective Context for the Safe Operation of Autonomous Systems

The safety of a system can only be demonstrated to have been achieved in a defined context. This is true whether it is a ‘traditional’ or autonomous system (AS). For traditional systems, a human is trusted to provide an oversight of operations, and react safely to unexpected scenarios that occur. For AS we cannot necessarily rely on human oversight to handle unexpected events, and must therefore be more confident that all possible hazardous scenarios are understood prior to operation. This makes the task of defining the context of safe operation (CSO) precisely and completely even more important for an AS so that unexpected scenarios can be limited. Attempting to define the CSO completely for an AS operating in a complex open-world environment could be an intractable task. It is therefore imperative that an effective and efficient way to define the CSO for AS can be found. Existing approaches to defining the CSO for AS are generally seen to be disjoint (in that each of the elements is considered and specified in isolation) and lacking in focus (in that the level of detail is found to be inconsistent and often inappropriate). What is required therefore is a targeted, iterative and integrated approach for defining the CSO for an AS. We provide an example of how this approach can be used to deliver an effective CSO for an autonomous robot.