Surrogate model framework for explainable autonomous behaviour in maritime robotic systems

With the surge in the application and deployment of robotic and autonomous systems across various sectors, there is an ever-growing demand for transparency and explainability. The maritime domain, which has embraced these technologies for tasks ranging from disaster site inspections to pipeline maintenance, is no exception. This chapter will discuss a novel surrogate model framework that aims to bring explainability to autonomous behaviours in maritime robotic systems, bridging the existing gap between complex robotic decisions and the human understanding of their actions with the desired mission intent. The chapter will provide an overview of why transparency and explainability are important for robotic and autonomous systems, especially in the maritime sector, exploring the complexity of conveying complex autonomous decisions in understandable terms for various stakeholders with potentially conflicting needs (e.g., in-field operators, remote controllers, and trained/untrained operators). It will introduce the framework of surrogate models for explainable decision-making, focusing on their utility in simplifying and approximating deterministic agent policies and offering explanations that are independent of the underlying autonomy model. Finally, the chapter will discuss the journey from simulated experiments to actual trials with maritime robots, emphasising the changes, challenges, and learnings experienced during this transition. The chapter will conclude by highlighting the potential growth areas, applications, and refinements for the surrogate model framework in the future.