Experience with Model-Based User-Centered Risk Assessment for Service Robots

Safety is now a major concern in many computer based systems and more particularly for autonomous systems such as service robots in physical contact with humans. The traditional approach to analyze the safety of such systems is to use risk assessment methods based on models of system structure, or system behavior. Unfortunately, such models are hard to produce for autonomous systems. We propose an approach based on the standardized risk assessment process which is applied during the initial phases of the development process. We first use the common Unified Modeling Language (UML) and a preliminary application domain hazard analysis without considering any robotic device. Then, during the specification phase, a risk assessment of the robotic system is carried out. It consists in modeling tasks in UML, identifying hazardous situations (including human errors), and estimating associated risks. We base this analysis on an adaptation of the guideword-based collaborative method HAZOP (HAZard OPerability) applied to UML models. The process has been successfully applied to the development of an assistive robot providing assistance for standing up, sitting down and walking, and health-state monitoring. Results in terms of integrability, usability, validity and applicability of the method are really encouraging. Majors benefits are a good management of the level of abstraction (and thus combinatory explosion is controlled), an easy communication between different stakeholders using basic UML diagrams, and a structured safety documentation required for certification.