Dynamic Behavior Risk Assessment for Autonomous Systems

Software-controlled technical systems are omnipresent in our daily lives. In many domains, such as automotive, avionics, and robotics, engineers are currently building systems that act without human assistance, in only partly defined open environments. These systems pose new safety challenges of which the challenge of unsafe nominal behavior is addressed in this work with a new approach for risk assessment: Dynamic Behavior Risk Assessment.

---

Software-controlled technical systems are omnipresent in our daily lives. In many domains, such as automotive, avionics, and robotics, engineers are currently building systems that act without human assistance, in only partly defined open environments. The most prominent example are self-driving vehicles. To achieve this automation in an open environment, technical systems need to be able to behave adequately in unthought-of situations. We refer to systems with this capability as autonomous systems.
Autonomous systems inevitably contain some degree of uncertainty regarding their behavior in such unthought-of situations. Consequently, it is not possible to analyze the full space of their behavior at development time. The current risk assessment approaches in Safety Engineering - the discipline responsible for creating systems with acceptable risk - rely on extensive analyses conducted during development time. This is not feasible for assessing the risk associated with the behavior of autonomous systems. This dissertation presents a novel approach for Dynamic Behavior Risk Assessment for autonomous systems to overcome this limitation.