Information-Theoretic Multi-Agent Path Planning for Risk-Aware Navigation Under Environmental Uncertainty

A. Nimmagadda, J. Park, A. Von Moll, I. Weintraub, and D. Casbeer

Published in Aviation Forum (Submitted for Review), 2026

Autonomous navigation in severe weather requires balancing information gathering and mission safety under extreme uncertainty. A novel multi-agent collaborative framework addresses this challenge through information-theoretic path planning that coordinates HighValue Aerial Assets (HVAA) with expendable Sensing Aerial Assets (SAA) in hurricane-like conditions. The framework employs modified Rapidly-exploring Random Tree star (RRT*) algorithms: Expected Travel Time RRT* for conservative HVAA planning and InformationTheoretic RRT* for SAA exploration that maximizes uncertainty reduction. To guide SAA behavior, an entropy-weighted utility function jointly optimizes travel cost and expected information gain, enabling targeted sampling of high-uncertainty regions characterized by bimodal wind distributions. A nonlinear trajectory-tracking controller with look-ahead guidance ensures feasible execution of planned path under strong wind disturbances. Hurricane Harvey 2017 wind data validates the approach, demonstrating 28% path length reduction and 24.1% travel time improvement compared to baseline non-collaborative planning. Under moderate wind conditions (6-8 m/s), the system achieves 2.0-meter average tracking accuracy with 32meter maximum deviation during aggressive maneuvers. Performance analysis reveals clear operational boundaries, with acceptable operation below 14 m/s wind speeds and graceful degradation beyond these limits. The framework enables safer autonomous navigation in uncertain environments while quantifying operational constraints for deployment guidance.