Risk-aware stochastic ship routing using constrained continuous belief tree search

Improved route planning for commercial shipping can enable reduced environmental impact, improve ship safety records, and lower fuel and maintenance costs. A fundamental challenge is to design ship routing algorithms that can contend with uncertain weather forecasts and real-world models of ship performance and safety. This paper introduces a stochastic ship routing framework that uses the conditional value-at-risk (CVaR) metric to guide the behaviour of a modified Continuous Belief Tree Search (CBTS) algorithm to find a safe and fuel-efficient long-haul shipping route. Our method provides a principled means to utilise a probabilistic representation of weather forecasts derived from ensemble forecasting for the purpose of route planning and allows for a user-defined threshold of risk tolerance. Another key advantage of our method is its ability to dynamically choose candidate route waypoints using weather-dependent estimates of fuel and safety information. Evaluation of long-haul routes through the Atlantic, Pacific and Indian oceans using real-world ship performance models and weather forecasts show considerable improvements in fuel usage and computation time compared to state-of-the-art ship routing algorithms. • This paper introduces and solves a novel stochastic ship weather routing problem. • The method finds low-fuel, low-risk, on-time routes through uncertain weather. • On real ship and weather data, our method uses 2.5-13% less fuel than others. • The method takes 5-30% less time to compute than other algorithms.