MERMAIDS - Marine Engine Robotic Maintenance and Inspection Development System

To support future missions for the US Navy using unmanned surface vessels (USVs), robotic systems will need to autonomously perform roles that would normally be accomplished by sailors.  One particularly important class of tasks to sustain operation status of a ship revolves around the maintenance of marine diesel engines.   This project will combine teams from HEBI Robotics and the University of Michigan to design a concept for, and perform initial validation of, a system for automated maintenance of marine diesel engines.  The system will include a combination of task-optimized robotic manipulators and tools that are capable of force and tactile feedback, integrated static and repositionable vision systems, and software algorithms that will be developed to allow a Phase II project to pursue fully autonomous capability. Engine maintenance includes a myriad of manipulation tasks that involve nuanced physical interaction with the world.  These include tasks such as: opening valves, rotating or pressing buttons and switches, tightening and untightening bolts, and servicing soft-flexible components such as gaskets and washers.  Despite recent advancements in learning from demonstration such tasks have continued to be difficult to automate and remain an open area of research.   There are also a number of inspection tasks that are critical to marine engine maintenance.  These include tasks like: checking for deformations, cracks, and leaks, recognizing and interpreting warning lights, and reading gauges or status message text.  While there is progress today of identifying anomalies for remote monitoring, autonomously turning these into actionable tasks of an autonomous system is similarly an open problem. The proposed project will address these technical challenges with a combination of robotic hardware and software development.  The hardware development will be led by HEBI Robotics, a small business that specializes in developing robotic components and systems for challenging manipulation and inspection tasks that lie outside the domain of traditional industrial automation.  The software development will be led by Dmitry Berenson and the Autonomous Robotic Manipulation Lab at the University of Michigan, whose lab specifically focuses on creating algorithms and learning methods that allow robots to interact autonomously with the physical world.