A Methodology for the Analysis of Robotic Systems via Process Mining

Robotic systems are widely adopted in various application scenarios. A very complex task for developers is the analysis of robotic systems' behavior, which is required to ensure trustworthy interaction with the surrounding environment. Available analysis techniques, like field tests, depend on human observations, while automated techniques, like formal analysis, suffer from the complexity of the systems. Recent works show the applicability of process mining for the analysis of event data generated by robots to increase the understanding of system behavior. However, robots produce data at such a low granularity that process mining cannot provide a meaningful description of the system's behavior. We tackle this problem by proposing a process mining-based methodology to prepare and analyze the data coming from the execution of a robotic system. The methodology supports the system developer in producing an event log compliant with process mining techniques and is used to analyze multiple perspectives of robots' behavior. We implemented the methodology in a tool supporting its phases. We use the tool on a robotic smart agriculture scenario to evaluate the feasibility and effectiveness of the methodology.