Single-objective voyage optimization — minimize fuel, or minimize time, or minimize emissions — is a solved problem. The commercial reality of shipping, however, is that every voyage involves simultaneous, conflicting objectives: the charterer wants minimum fuel, the owner wants maximum safety margin, the compliance officer wants CII improvement, and the master wants a route she can actually trust.
We present a multi-agent system architecture where five specialized agents — Route Generator, FOC Estimator, Emissions Agent, Safety Agent, and Trade Agent — each optimize for their primary objective and negotiate through a structured communication protocol. The negotiation is formalized as a cooperative game, with a game-theoretic optimizer resolving conflicts to produce a Pareto-optimal route given the operator's stated priority weights.
Each agent's reasoning is transparent and auditable: the Route Generator explains its waypoint choices in terms of weather windows and great-circle deviations; the Safety Agent flags specific weather conditions that triggered a detour; the Emissions Agent quantifies the CII impact of each route option.
In simulation over 500 multi-objective voyage scenarios, the multi-agent system achieved a 4.3% improvement in the combined objective function versus the best single-agent baseline, while maintaining full interpretability of the final route recommendation.