A Resilient Positioning Method via Active Fault Detection for Unmanned Vehicles With Low-Redundancy Sensors

Abstract

Continuous and reliable positioning is essential for unmanned vehicles operating in adverse or hostile environments. However, practical constraints on power, space, and payload often limit sensor redundancy, making positioning systems more vulnerable to sensor faults under harsh conditions. Robust fault detection and identification (FDI) are therefore critical to resilient positioning, but existing methods often struggle in scenarios with low sensor redundancy and concurrent sensor faults. In addition to conventional sensor cross-checking methods, we introduce a control signal-based active fault detection strategy in which sensors are sequentially integrated into a closed-loop control system to track a predefined trajectory. The feedback error and its rate of change are monitored to assess the sensor health. Diagnostic thresholds are derived using a Lyapunov function and the state covariance matrix. The proposed framework enables effective fault detection without requiring extensive sensor redundancy or an accurate mapping between control inputs and sensor outputs. Simulations on an autonomous drone demonstrate that, under concurrent multifault conditions, the proposed method achieves a correct detection rate exceeding 81 %, outperforming the best-performing baseline method by 10.8 percentage points. This improved fault identification capability leads to a 62.5 % reduction in positioning root mean square error (RMSE) compared with the best baseline. Real-world vehicle experiments further validate the method, demonstrating accurate isolation of faulty sensors when two of the three sensors malfunctioned simultaneously, with RMSE bounded within 1.06 m and the 95th-percentile error reduced by 47.6 % compared with the best-performing baseline.