Automated extraction and checking of property models from source code for robot swarms

Ettore Merlo, Carlo Pinciroli, Jacopo Panerati, Michalis Famelis, Giovanni Beltrame

Abstract: As robots become a common presence in our everyday lives, ensuring the security and safety of robotic systems becomes an increasingly important and urgent challenge. Multi-robot systems, in particular, have the potential to revolutionize multiple industries---such as transportation and home care---where safety guarantees are a primary requirement. A known challenge for swarms and multi-robot systems is the gap between requirements and design, due to the need to translate swarm-level objectives into robot-level behaviors. In this paper, we focus on a less studied problem---the gap between requirements and implementation. As a case study, we use Buzz, that is a dynamic programming language designed for swarm robotics applications. Similarly to Python, Lua, and JavaScript, Buzz does not natively offer formal guarantees of correctness or safety. We propose an approach to automatically extract "as-implemented" models from Buzz programs, whose properties can then be formally analyzed and verified. Results obtained from the experiments performed on two medium-size open-source production-level systems for robotics research have also been reported. Our results show that the approach is feasible and is scalable to larger systems.