Abstract

In the marine research fleet of the future, untethered autonomous vehicles will be much more prevalent than they are today, meeting the rapidly-growing demand for environmental data to inform policy decisions relating to climate, sea level rise and ecosystems. This means that controllable autonomous vehicles will need to be used optimally to perform data gathering tasks, especially in remote and hostile regions. To fully utilise the autonomous fleet, there will be a need for intelligent automated decision-making to command and control its role in long-term research campaigns. Instead of each vehicle being separately tasked and piloted, as is the case now, an automated strategic planning system will coordinate the fleet to maximise the ocean area covered and optimise the quantity, quality and type of data collected over the lifetime of a research campaign. Strategic plans can be used to augment, and eventually even replace, some of the human expertise currently needed to manage the fleet.

To generate high value plans for the fleet, a strategic decision-making system will need to be informed by environmental simulations and predictions based on data. It will need an understanding of the risk to the vehicles, and how to minimise this risk without wasting opportunities. This talk will describe a plan-based approach to coordinating the activities of a fleet of autonomous vehicles undertaking long-term campaigns in the Southern Ocean. The talk will discuss how predictions made by Machine Learning can be integrated with combinatorial decision-making to generate plans that use the fleet safely and effectively.