Marine Autonomous Vehicle Fleet Scheduling to Maximise Scientific Impact

Marine science increasingly depends on continuous, high-resolution data to understand oceanic processes and climate impacts. Traditional research vessels, though reliable, are costly, slow, and limited in coverage. The advent of MAVs offers a flexible, energy-efficient alternative capable of operating in harsh environments, including polar regions and deep seas. Prior research has focused on path planning algorithms like A* and sampling-based methods, but scaling these to large fleets remains a challenge. Recent efforts incorporate multi-vehicle coordination, yet often neglect integration with maritime logistics, such as vessel support and dynamic re-routing. As fleet sizes expand, the need for comprehensive, scalable optimization models becomes critical. Existing heuristic approaches lack guarantees of optimality and struggle with real-time adjustments. This paper builds on this foundation, proposing a MILP-based solution that unifies vessel and vehicle scheduling, addressing energy constraints, operational constraints, and logistical support in a single framework, thus advancing the state of the art in autonomous marine fleet management.

The core challenge lies in efficiently scheduling large-scale MAV fleets to maximize scientific data collection within strict energy, time, and logistical constraints. The problem involves multi-objective optimization: balancing data volume, energy consumption, fleet size, and operational costs. Additional complexities include integrating vessel support for battery swaps and transit, handling environmental variability, and ensuring real-time adaptability. The combinatorial nature of route assignment, task scheduling, and resource management results in a computationally intensive problem, especially at scales involving hundreds of vehicles and multiple support ships. Existing methods lack the capacity to solve such large, multi-constraint problems efficiently, necessitating advanced optimization techniques capable of providing near-optimal solutions within practical timeframes.

Key innovations include: 1) embedding traditional ship itineraries into a MILP framework, enabling joint vessel and MAV scheduling; 2) supporting mid-mission battery swaps and transit via support ships, increasing operational flexibility; 3) designing a multi-objective function that balances data maximization, energy efficiency, and fleet minimization; 4) employing advanced branch-and-bound algorithms for rapid solution generation at large scales. These innovations address critical gaps in existing literature, which often treat vehicle path planning and maritime logistics separately. By integrating these elements, the model offers a holistic, scalable solution that aligns with real-world operational demands, facilitating large-scale, cost-effective marine observation missions.

• �� Define the problem scope, including research stations, MAVs, ships, and tasks, with relevant spatial-temporal parameters. • Formulate a multi-objective MILP model, incorporating variables for route selection, task assignment, battery levels, and ship support. • Establish constraints for energy limits, time windows, task requirements, and vessel schedules, ensuring feasibility. • Implement solution algorithms combining branch-and-bound with heuristics to handle large problem sizes efficiently. • Support mid-mission battery swaps and transit via support ships by adding dedicated decision variables and constraints. • Validate the model through extensive simulations, testing different fleet sizes, environmental scenarios, and parameter settings. • Generate visualizations and scenario analyses to facilitate stakeholder understanding and strategic planning.

Simulations used synthetic ocean environments with varying fleet sizes (10-200 MAVs), task complexities, and environmental conditions. Baseline comparisons included heuristic and greedy algorithms. Metrics evaluated were total data volume, energy consumption, computation time, and solution robustness. Hyperparameters such as objective weights and time window widths were tuned via sensitivity analysis. Multiple runs assessed solution consistency and scalability. Results demonstrated that the MILP model consistently outperformed baselines, solving large problems within minutes and achieving 20-30% higher data collection efficiency, validating its practical applicability for real-world deployments.

The model efficiently solved scheduling problems for 30 MAVs in under 5 seconds, scaling to 100 MAVs within 3-4 minutes, with solution quality surpassing heuristic methods by over 20%. Data collection increased by approximately 22%, while energy consumption decreased by 15%. Support for mid-mission battery swaps and ship-assisted transit contributed to these improvements. The model's scalability was confirmed through consistent performance across scenarios, demonstrating its potential for operational deployment in large-scale marine observation missions. Visualizations clarified complex schedules, aiding stakeholder understanding and decision-making.

This framework is directly applicable to polar research, deep-sea exploration, and environmental monitoring projects requiring extensive, coordinated data collection. It enables agencies and companies to plan efficient, cost-effective autonomous surveys, reducing reliance on traditional vessels. Future integration with real-time sensor data and adaptive algorithms will allow dynamic re-scheduling, further enhancing operational resilience. The approach can also inform policy-making for sustainable ocean resource management by providing reliable, high-resolution data streams.

The model assumes accurate environmental and vessel schedule data; uncertainties could affect solution feasibility. Computational costs increase with fleet size and scenario complexity, potentially limiting real-time responsiveness. Sensor fusion, task prioritization, and adaptive re-planning are not fully addressed, which may limit flexibility in rapidly changing conditions. Further research is needed to incorporate stochastic environmental models and real-time data assimilation for operational robustness.