Uncertainty-Aware Velocity Correction for Proprioceptive Vehicle Localization using Evidential Mamba

2026-07-06

AI Summary

This work addresses the challenge of positioning failure in GNSS-denied environments caused by inertial measurement unit (IMU) drift by proposing a real-time, high-precision localization method that requires no additional hardware. The approach leverages the Mamba selective state space model to capture the temporal dynamics of vehicle motion and integrates evidential deep learning—based on a Normal-Inverse-Gamma distribution—to generate virtual velocity observations with quantified uncertainty. These uncertainty-aware velocity estimates are then fused into an error-state extended Kalman filter to correct IMU drift. This study presents the first integration of Mamba with evidential learning for uncertainty-aware velocity estimation. Experimental results on real-world vehicular data demonstrate that the method achieves localization accuracy within 90% of that attainable with dedicated external speed sensors and supports real-time deployment at 40 Hz on edge devices, effectively mitigating prolonged GNSS outages.

Abstract

Reliable localization in GNSS-denied environments remains a fundamental challenge for intelligent vehicles, as inertial navigation systems accumulate unbounded drift without external correction. Existing approaches provide drift correction through dedicated infrastructure, expensive external sensors, or complex multi-sensor fusion, each introducing practical deployment barriers. We propose Evidential Velocity Correction using Mamba (EVC-Mamba), a learning-based architecture that transforms onboard vehicle sensor data into a virtual velocity sensor for IMU drift correction without additional hardware. A Mamba-based selective state space model captures the temporal dynamics of vehicle motion, while evidential deep learning with a Normal-Inverse-Gamma distribution provides principled uncertainty quantification. The resulting uncertainty-aware velocity estimate is incorporated as a virtual correction measurement into an Error-State Extended Kalman Filter to reduce position drift. Evaluation on real-world vehicle data demonstrates that inertial navigation using the proposed velocity correction achieves localization accuracy within 10% of a dedicated external velocity sensor across different outage durations. The proposed architecture supports real-time onboard deployment at 40 Hz on edge hardware, enabling reliable localization during prolonged GNSS outages.

Problem

• GNSS-denied localization
• inertial navigation drift
• vehicle localization
• velocity correction
• uncertainty quantification

Innovation

• Evidential Deep Learning
• Mamba
• Velocity Correction
• Uncertainty Quantification
• GNSS-denied Localization