Towards a Cognitive Architecture for Marine Robots

In this paper, we introduce a proven cognitive architecture to the underwater domain, where such architectures are becoming increasingly needed. Components of the proposed framework include a mission planner and executioner, knowledge base and reasoner, real-time photorealistic simulator, web-based service for replaying and explaining past missions, and ensemble-of-experts perception system. Similar isolated cognitive components have already proven useful in marine robotics. What sets the proposed framework apart is that it offers the potential for a complete, consistent, end-to-end solution—from sensor perception and actuator control to task selection and plan adaptation to post-mission analysis and explainability. The same components proposed for this framework have already found success in settings ranging from both personal and retail service robotics to collaborative human-robot search-and-rescue missions. We present the most significant additions and adaptations to these components in order for them to prove equally apt for underwater robots. To guide the discussion, we consider a motivational scenario involving a pipeline inspection performed by an autonomous underwater vehicle.