SENATOR Working Report "Needs Exploration for Long-Term Autonomous Marine Systems" surpasses 250 downloads

The working report on long-term autonomy needs published last summer has surpassed 250 downloads. Combining crew interview, state-of-the-art surveys, and STPA-inspired analysis, the report traces research needsfor long-term autonomous vessels. You can get the report online here

Executive Summary

Making a marine platform autonomous for long-duration missions of weeks to months at sea has been shown to be a fundamentally different challenge than those faced by crewless air and ground vehicles. Self-health assessment, mission planning and logistics consideration are all heightened for the marine vessel that may be out for weeks to months at a time. How- ever, how such assessments and mission planning is done on crewed vessels today is not well understood, which makes researching algorithms for this area difficult.

This work reports on a broad framework of risk concerns for such long-term missions, and then reviews current marine systems in service. Existing platforms show a clear tradeoff between platform complexity and achievable endurance. Gliders and simple platforms have completed weeks-to-month voyages but with high loss rates. More complex vessels are still in the days-to-low-weeks range of mission lengths, and no vessel currently performs long-term planning autonomously. A new three-component rating system was proposed to track plat- forms, using the platform’s decision-making, endurance, and platform complexity as metrics. Existing platforms were visualized with this system, confirming the capability gap.

A series of interviews with ship and shore crews was used to attempt to determine how such planning is done today. This process revealed that machinery systems comprise the central concern around platform health. The maintenance of machinery is highly structured within the vessel’s preventative maintenance system (PMS). However, both integrating the overall health of the platform, and weighing risks for planning were done in a human-center manner. Such planning was not standardized or recorded in a formal procedure, but used extensive human experience and implicit criteria. Shore-side support was also widely used in diag- nosing problem, planning repairs, and supporting on-board decisions. Thus, the interviews produced a list of concerns but not a definite planning approach that could be automated.

A high-level simulation-based approach was used to see how accurate planning around plat- form health had to be to improve operational performance. A fleet of 10 vessels was used to maintain a patrol line at different distances from a support base, with each vessel having a single health parameter that degraded stochastically. Four different planning approaches were compared, and the result showed that even imperfect long-term planning systems may produce large gains in platform effectiveness vs. static rule-based approaches.

Finally, a modified STPA approach was used to try to explore significant risk areas for long-term planning systems. Two STPA formulations were compared, and the approach more narrowly focused on mission planning was able to identify broad areas where existing algorithms may be insufficient. A table of resulting challenge areas was constructed, and three development case studies were proposed to address the gaps in the table. These cases were designed to be tractable for basic research exploration yet involving enough disciplines to be broadly representative of the at-sea mission planning problem. The three case studies included a fuel management study, a machinery design and support case, and an adaptable risk level case. Suggestions for implementing these case studies, and further work, finalize the report.

Matthew D. Collette
Matthew D. Collette

My research interests include machine learning in design, data-model fusion, and lifecycle analysis