Overview
FAST-LIVGO (LiDAR-Inertial-Visual-GNSS Fusion Odometry) is a proposed framework designed to enhance robust state estimation and mapping in challenging robotic environments. The system aims to overcome limitations of existing multi-sensor fusion approaches, particularly in scenarios characterized by degeneracy or high dynamics. It integrates LiDAR, Inertial Measurement Unit (IMU), visual, and Global Navigation Satellite System (GNSS) data within a tightly coupled architecture.
Research Context
Robust state estimation and mapping are fundamental challenges in robotics, especially in environments that are long-term, large-scale, and highly dynamic. Prior LIVO (LiDAR-Inertial-Visual Odometry) systems exhibit strong local accuracy but are susceptible to accumulated drift over prolonged operations. These systems may also experience failure in scenes with geometric degradation or a lack of texture. GNSS-aided fusion frameworks often depend on LiDAR or visual odometry for state prediction and outlier rejection. This dependency makes them vulnerable to performance degradation when the underlying odometry systems encounter degeneracy.
Approach
The FAST-LIVGO framework employs a tightly coupled LiDAR-Inertial-Visual-GNSS fusion approach, built upon an Error-State Iterated Kalman Filter. Key elements of the approach include:
- Online Spatiotemporal Alignment: A module leveraging Dynamic Time Warping (DTW) explicitly addresses highly dynamic conditions by performing online spatiotemporal alignment.
- Enhanced GNSS Precision Exploitation: The system develops observation models to better utilize GNSS precision. These models are based on Doppler shifts and fixed-anchor Time-Differenced Carrier Phase (TDCP). This allows for millimeter-level relative constraints without the need to augment historical anchor states within the filter.
- Degeneracy-Aware Dual-Mode Outlier Rejection: A strategy for outlier rejection is designed to be aware of the LIVO system's degeneracy level. This strategy enables switching between two modes:
- LIVO-prior-guided rejection, used when LIVO performance is adequate.
- GNSS-aided recovery, activated when LIVO degeneracy is detected, to restore accurate state estimation.
Findings
Experiments were conducted using two distinct datasets: the public M3DGR dataset and a custom dataset collected by a fixed-wing UAV operating at 20 m/s. The evaluation indicated that the FAST-LIVGO system, when compared to state-of-the-art methods:
- Reduces accumulated drift.
- Minimizes map ghosting effects.
- Demonstrates improved accuracy.
- Exhibits enhanced robustness.
Why This Matters
The findings suggest that FAST-LIVGO offers advancements in robust state estimation and mapping. By reducing accumulated drift and map ghosting, and by demonstrating improved accuracy and robustness, the framework could enhance the reliability of autonomous systems operating in complex and dynamic environments over extended periods. The ability to manage instances of odometry degeneracy represents a potential step forward for robotic navigation systems.