Autonomous fleet ice management via reinforcement learning
Dr. Andrew Vardy, Memorial University of Newfoundland, Computer Science & ECE (Joint)
Dr. Kevin Murrant, Research Council Officer / Ocean, Coastal and River Engineering
Ice management is a complex and expensive undertaking that generally requires special purpose icebreaking vessels. This project will investigate whether a fleet of autonomous vessels, anticipated to be much smaller than typical icebreakers, can manage ice fields by pushing the ice away from the channel of interest. The autonomous nature of this fleet is crucial as it would allow operation over a larger time window than conventional vessels. In principle, ice management could become a continuous process, greatly easing the planning process for commercial vessels to utilize the waterway.
Rather than applying a traditional model-based approach, this project will employ existing ice simulation technologies developed at NRC-OCRE in combination with model-free reinforcement learning to arrive at a design for the vessel controllers that maintains the channel’s navigability.
- The desired skill set for the student would be built on the strong foundation of an undergraduate degree in either computer science, computer engineering, or a related discipline
- It is expected that the student would also have a Master’s degree where they obtained conceptual exposure and practical skills in the disciplines of robotics and machine learning
- Practical skills in conducting physical experiments and dealing with real-world experimental data would also be crucial to experiments conducted within the NRC-OCRE’s facilities—particularly the ice basin
Position is still open.