Dynamic Weighted Multi-Sensor Fusion for Mobile Robot Localization Using the Error-State Kalman Filter

A dynamic weighted multi-sensor fusion method is proposed to address localization degradation in four-wheeled differential mobile robots under GNSS denial and wheel slippage conditions. Sensor data from the IMU, wheel odometer, and RTKGNSS are first preprocessed. A dual-threshold decision mechanism, based on RTK confidence evaluation and wheel speed-IMU displacement residual detection, is then constructed to determine system states. Subsequently, observation noise weights for each sensor are dynamically adjusted using a noise-driven model and incorporated into an error-state Kalman filter (ESKF) for position estimation. Experimental results demonstrate that, under GNSS denial or wheel slippage, the proposed method effectively mitigates the impact of partial sensor failures and significantly enhances the robustness of the localization system against dynamic disturbances.