Comparison of AI Planning frameworks for underwater intervention drones

Abstract

Increasing the autonomy level of underwater robots is bound to bring significant benefits to subsea operations in terms of cost reduction, efficiency and safety. Artificial Intelligence (AI) planning is an important component of autonomy, since its objective is to generate a feasible and efficient sequence of high-level actions (plans) that are subsequently executed by the control system. Despite being an important part of space missions, AI planning has not been given much attention in underwater applications, with a few notable exceptions. In this paper we study two AI planning frameworks, ROSPlan and T-REX, and implement them to a subsea scenario formulated in collaboration with the industry. The two frameworks tackle the sense-plan-act paradigm in different ways, with the T-REX demonstrating a reactive behavior that can be advantageous to subsea operations, especially given their dynamic and uncertain nature.