Bridging Process Mining and Robotic Systems

As robotic systems become increasingly integrated into real-world systems, the need to enhance our understanding of their operations continues to grow. Bridging the gap between the process mining discipline and robotics can improve the transparency, accountability, and efficiency of these systems. However, the application of process mining techniques is hindered by the nature of robotic event data, which is mainly fine-grained, such as sensor readings, rather than high-level activity events. In this context, this exploratory paper reports on the collection and classification of publicly available robotic datasets. We identify a collection of 118 datasets that we classify based on characteristics like application domain, type of involved robot, onboard sensors, collaborative capabilities, and readiness for processing with process mining techniques. We also assess the application of well-known artificial intelligence techniques for preparing fine-grained robotic data and obtaining high-level activities. In particular, we use a statistical model on position data and the fine-tuned vision foundation model for videos. Building on these findings, we highlight challenges and opportunities in applying process mining to robotics to create a holistic understanding of their operations.