Skill-Based Architectures in Autonomous Systems: an Empirical Study

The deployment of complex autonomous systems in open environments requires robust, modular, and verifiable software architectures. Skillset models, which encapsulate robot capabilities into modular Skills, address these challenges. In such architectures, the Skill-based layer translates high-level plans into executable actions. Using a Domain-Specific Language (DSL) for Skillset modeling enables formal verification and reliable code generation.This paper presents an empirical study of recent Skill-based architectures using the ONERA Robot Skill Toolchain, highlighting best practices, common pitfalls, and guidelines for future developments.Our results confirm that Skill-based architectures improve adaptivity, complexity management, and robustness in autonomous systems. However, ensuring dependable autonomy still requires formal development frameworks and better failure-handling tools.