Abstract

Software architectures for autonomous systems are generally structured with 3 layers: a decisional layer managing autonomous reasoning, a functional layer managing reactive tasks and processing, and an executive layer bridging the gap between both. The executive layer plays a central role, as it links high-level tasks with low-level processing, and is generally responsible for the robustness or the fault-tolerance of the overall system. We will present  a development process for such an executive layer that emphasises on the dependability of this layer. To do so, we structure the executive layer using skills, that are formally defined using a specific language, and we then provide some tools to verify these models, generate some code, and a methodology to assess the fault-tolerance of the resulting architecture. The decisional layer is then agnostic of the skillset layer, and can be implemented in various ways. We will present some examples of applications in the field of mobile autonomous robots that embeds such three-layered skill-based architecture. Autonomous Planning is then a central aspect of the decision-making aspects of the robotic missions.