Physiological Signal-Guided Uncertainty Management for Autonomous UAVs in Human–UAV Supervisory Control

This paper presents a physiological signal-guided uncertainty-management framework that integrates real-time indicators of the operator’s supervisory state into UAV autonomy to improve obstacle avoidance and risk-aware decision-making under uncertain conditions. During UAV supervisory control, multimodal physiological signals, including heart rate variability, blood pressure, electrodermal activity, and other cardiovascular or stress-related measures, are time-synchronized with UAV telemetry, perceived obstacle fields, planner confidence, environmental uncertainty estimates, and operator intervention logs, including waypoint edits, overrides, and replanning commands. These heterogeneous data streams are fused using a Bayesian hierarchical state-space framework to estimate latent supervisory states representing trust miscalibration, risk sensitivity, and situational-awareness degradation. The estimated states are then incorporated into the UAV decision-making stack as bounded uncertainty-management parameters that regulate safety margins, replanning priority, and risk preference without relaxing hard safety constraints. The framework was evaluated in a simulation-based human-in-the-loop study involving 24 operators and 180 UAV obstacle avoidance missions. Performance was assessed using held-out-operator mission success AUROC, proxy-state RMSE, mission success rate, mean minimum obstacle clearance, mission risk index, operator override rate, and replanning latency. The results support the feasibility of using physiological, behavioral, vehicle, and environmental information to adapt UAV supervisory control to uncertainty in both the operating environment and human supervisory readiness.