Analysis of human errors in human-autonomy collaboration in autonomous ships operations through shore control experimental data

Human-autonomy collaboration plays a pivotal role in the development of Maritime autonomous surface ships (MASS), as Shore control center (SCC) operators may engage in the control loop by directly operating the MASS, or, in the supervisory loop, monitoring the MASS and taking over control when needed. Thus, efficient human performance during takeover control and operation is crucial for the safety of MASS operations. However, since the MASS is still in the early phase of development, the mechanism of human errors is unknown, and the data on human-autonomy collaborative operation is scarce. Human reliability analysis (HRA) aims to assess human errors qualitatively and quantitatively, and is widely used in various complex systems to help safety analysis. This study is dedicated to incorporating advanced HRA methods elements to identify and quantify human errors during taking over control and operation of a MASS in collision avoidance scenarios. It presents virtual experimental results, combined with theoretical human error identification and assessment methods. At first, we apply the Human-System Interaction in Autonomy (H-SIA) method to identify potential human errors; secondly, we identify relevant Performance Shaping Factors (PSFs) including Experience, Boredom, Task complexity, Available time and Pre-warning, and performance measures of the human errors, and implement them in the virtual experiment based on a full-scale autonomous ferry research vessel called milliAmpere2. Finally, we build a Bayesian Network (BN) to present causal and probabilistic relationships between PSFs and human errors through experimental data. The results show that available time has the highest impact on takeover performance of operators, followed by task complexity and pre-warning. Boredom does not present a significant sole impact unless combined with available time. Experience does not show a significant impact on human performance. In addition to the relevance of the human errors analysis to the safe development and operational design of MASS, the developed method benefits other human-autonomy collaborative systems. The developed BN model shows adaptability to assess human error probabilities, and the practical significance of integrating experimental data into the existing HRA methodologies for complex systems.