A skill fault model for autonomous systems

Autonomous systems are now deployed for many applications to perform more and more complex tasks in open environments. To manage complexity of their control software architecture, a current trend is to use a 3-layers approach, with a decisional layer (able to formulate decisions), a functional layer (low level control actions), and between them a skill layer. This layer is dedicated to convert high level plan objectives into low level atomic actions, sent to the functional layer. In order to deal with failures that may happen at runtime, detection mechanisms and reaction strategies may be implemented in these layers, or even in external devices. However, no generic technique is available to guarantee that all these mechanisms will be consistent. We present in this paper an approach that focus on the skill layer, with a proposal of a generic skill fault model used to design and analyze failure detection and reactions mechanisms. This approach has been successfully applied to a real drone application, and we present an extract of the resulting fault analysis models.