TL;DR
This paper tackles the challenge of maintaining accurate vehicle positioning when Global Navigation Satellite Systems (GNSS) like BeiDou are unavailable, a common issue in urban environments. They propose a novel fusion system combining a Micro-Inertial Measurement Unit (MIMU) with an odometer and vehicle state constraints (MO-C). This approach significantly improves positioning accuracy and reduces errors during GNSS outages, offering a robust solution for autonomous driving.
What Problem Does This Paper Address?
Autonomous vehicles rely heavily on precise and continuous positioning information. While GNSS (like GPS or BeiDou) and Inertial Navigation Systems (INS) are standard, GNSS signals are frequently blocked in dense urban areas by tall buildings. This leads to poor satellite geometry, signal loss (NLOS/multipath), and ultimately, inaccurate positioning. INS, while providing continuous data, suffers from rapidly accumulating errors without external correction. Existing solutions often focus on improving INS noise suppression or integrating additional sensors, but this paper specifically addresses the challenge of maintaining accuracy during GNSS outages by fusing MIMU and odometer data with novel state constraints.
Methods
The core of this work is the development of a Micro-Inertial Measurement Unit (MIMU)/Odometer fusion system with vehicle state constraints (MO-C).
- MIMU/Odometer Fusion (MO): This forms the baseline fusion approach, combining the inertial data from the MIMU with distance measurements from an odometer. This helps to mitigate the rapid divergence of INS errors by providing a measure of distance traveled.
- Vehicle State Constraints (M-C): The authors incorporate "non-holonomic constraints" (NHC) and other vehicle kinematic constraints. These constraints leverage the natural motion of a vehicle on a road – specifically, that its velocity in the lateral (X) and vertical (Z) directions within its own body frame is typically near zero when moving normally. These constraints are used to suppress INS errors.
- MO-C System: This is the proposed integrated system that combines the MIMU/Odometer fusion with the vehicle state constraints. The system uses a Kalman filter to fuse the data and estimate the vehicle's state (position, velocity, attitude) while accounting for sensor biases and noise.
The paper details the mathematical models for the state equations and measurement equations for both the loose integration and the constrained fusion approaches.
Datasets
The paper does not mention specific publicly available datasets. Instead, it describes two field tests conducted by the authors:
- First Field Test: MIMU and Odometer data were collected and then post-processed using software implemented in MATLAB.
- Second Field Test: The algorithm was implemented in real-time using a DSP+FPGA hardware platform, and the results were obtained from real-time processing.
The tests involved a vehicle equipped with the described sensors, and the data was collected during actual driving scenarios, including simulated GNSS signal outages.
Key Insights and Ideas
- Odometer for Forward Motion, Constraints for Height: The odometer proved highly effective at reducing latitude and longitude errors, essentially tracking the vehicle's forward movement. The state constraints, on the other hand, were crucial for suppressing height errors.
- Heading Angle is Critical: The accuracy of the navigation solution is heavily influenced by the heading angle. If the heading angle can be determined precisely, the MO-C system can provide long-term, accurate 3D navigation solutions even without GNSS.
- Synergy of Odometer and Constraints: The combination of odometer data and state constraints (MO-C) yielded the best results, significantly outperforming individual sensor fusion or constraint-only approaches.
- Real-World Validation: The paper demonstrates the effectiveness of the MO-C system through both post-processing and real-time field tests, including scenarios simulating GNSS signal outages.
Key Results
The paper presents quantitative results from its field tests, comparing the performance of MIMU, MO, M-C, and MO-C.
- Post-Processing (90s test):
- Latitude/Longitude Errors: MO and MO-C reduced latitude errors by ~98.8-98.9% and longitude errors by ~95.1-95.5% compared to MIMU alone.
- Height Errors: MO-C reduced height errors by 95.9% compared to MIMU alone.
- Velocity Errors: MO-C achieved significant improvements in east velocity (87.2%), north velocity (96.9%), and especially up velocity (99.2%) compared to MIMU.
- Real-Time (GNSS Outage Simulation):
- M-C (Constraints only): Showed improvement over MIMU but still had significant errors, particularly in heading.
- MO (Odometer only): Stabilized latitude and longitude errors but showed increasing up-velocity errors.
- MO-C (Odometer + Constraints): Achieved the best performance, with 90-second position errors for latitude, longitude, and altitude within a few meters (e.g., latitude error ~1.72m, longitude ~-1.36m, altitude ~-4.38m). Velocity errors were also significantly suppressed (e.g., up velocity error ~-0.23m/s).
- During GNSS Outage (BDS/MIMU integration): The MO-C system maintained position and velocity errors within 10m and 0.2 m/s respectively, even when GNSS signals were completely lost for extended periods.
Limitations and Future Work
- Heading Angle Accuracy: The paper acknowledges that the accuracy of the navigation solution is heavily dependent on the heading angle. Errors in heading can lead to divergence over time, especially in long-term GNSS-denied scenarios.
- Testing Duration: The primary tests were conducted over 90 seconds. While promising, longer-duration GNSS outages might reveal further error accumulation.
- Urban GNSS Availability: The experiments simulated complete GNSS outages by removing the antenna. In real urban environments, some satellites might still be visible, albeit with poor geometry. The paper suggests that these remaining satellites could potentially aid navigation solutions.
- Parameter Calibration: The authors suggest that when GNSS is available, parameters for MIMU and the odometer could be estimated and calibrated, which would further improve performance during outages.
- Further Investigation: The authors propose further investigation into methods for obtaining better heading angle estimates, utilizing partially available GNSS signals in urban areas, and leveraging GNSS availability for sensor calibration.
Summary updated