Robert Emmitt

Robert Emmitt is an autonomy and data science engineer at PacMar Technologies with over seven years of experience. He holds a bachelor's and master's degree in naval architecture and marine engineering from the University of Michigan. In the last three years, Robert has focused on enhancing perception and navigation. Currently, Robert is a key contributor to three projects focused on advanced technology and shipboard digital twin incorporation. Prior to his current role, he was a member of the Concept Formulation (CONFORM) team at General Dynamics Electric Boat.

---

Presentation

Zero-shot learning for enhanced maritime navigation: a semantic reinforcement approach

This effort focuses on enhancing deep learning networks by adapting semantic relations and concepts in a knowledge graph to ground emerging categories from perception. The goal is to advance a maritime platform’s capabilities when encountering novelty and enable adaptive behavior in unstructured environments by providing it with a perception-knowledge-based toolset for improved navigational performance. The approach involves knowledge adaptation via reinforcement learning to improve the classification of previously unseen objects using both naval-specific initial knowledge and monocular vision. This is achieved by integrating conventional methods with commonsense knowledge bases. Results demonstrate improvements in traditional classification and novel object detection.