Robust and Precise Vehicle Localization based on Multi-sensor Fusion in Diverse City Scenes (1711.05805v2)

Abstract: We present a robust and precise localization system that achieves centimeter-level localization accuracy in disparate city scenes. Our system adaptively uses information from complementary sensors such as GNSS, LiDAR, and IMU to achieve high localization accuracy and resilience in challenging scenes, such as urban downtown, highways, and tunnels. Rather than relying only on LiDAR intensity or 3D geometry, we make innovative use of LiDAR intensity and altitude cues to significantly improve localization system accuracy and robustness. Our GNSS RTK module utilizes the help of the multi-sensor fusion framework and achieves a better ambiguity resolution success rate. An error-state Kalman filter is applied to fuse the localization measurements from different sources with novel uncertainty estimation. We validate, in detail, the effectiveness of our approaches, achieving 5-10cm RMS accuracy and outperforming previous state-of-the-art systems. Importantly, our system, while deployed in a large autonomous driving fleet, made our vehicles fully autonomous in crowded city streets despite road construction that occurred from time to time. A dataset including more than 60 km real traffic driving in various urban roads is used to comprehensively test our system.

• The paper introduces a multi-sensor fusion approach achieving centimeter-level accuracy in diverse urban environments.
• It integrates LiDAR intensity with altitude cues alongside GNSS RTK and IMU data to overcome individual sensor limitations in adverse conditions.
• Experimental results demonstrate an RMS error of 5-10 cm, marking a significant improvement over previous localization methods.

An Analysis of Multi-Sensor Fusion for Vehicle Localization in Urban Environments

The paper "Robust and Precise Vehicle Localization based on Multi-sensor Fusion in Diverse City Scenes" focuses on advancing vehicle localization technologies necessary for autonomous driving systems. It introduces a system that leverages multi-sensor fusion, incorporating GNSS, LiDAR, and IMU, to attain centimeter-level accuracy in localization, even within challenging urban landscapes, such as city downtowns, highways, and tunnels. This research is grounded in the context of autonomous vehicles from the Baidu Autonomous Driving Business Unit, enhancing the precision and robustness of localization while managing the typical challenges posed by urban scenarios.

Overview of the Methodology

The authors propose an innovative approach by not solely relying on LiDAR intensity or 3D geometry but using LiDAR intensity combined with altitude cues. This integration significantly enhances the system's accuracy and robustness, especially in scenarios with road construction or adverse weather conditions, where traditional LiDAR methods might falter.

1. Multi-Sensor Fusion Framework: The paper employs a multi-sensor fusion strategy that adaptively combines data from GNSS Real Time Kinematic (RTK) systems, LiDAR, and IMUs. This integration allows the system to compensate for the shortcomings of individual sensors. For instance, while GNSS RTK offers high precision, it is prone to errors in environments with signal blockage. In contrast, LiDAR is affected by weather conditions but thrives in environments rich in 3D features.
2. LiDAR-Based Localization: The methodology includes a robust LiDAR localization algorithm using a pre-generated map that maintains a grid-cell representation of laser intensity and altitude landmarks. LiDAR data is processed through a histogram filter enabling discrete state-space decomposition for optimal localization results.
3. GNSS-RTK Module and Ins-Aided Ambiguity Resolution: The GNSS component of the framework incorporates the RTK module, which is supported by INS, mitigating multipath and signal blockage issues common in urban settings. The system employs an error-state Kalman filter, enhancing the fusion framework's versatility and handling diverse urban scenes competently.

Numerical Results and Performance Evaluation

The system's efficacy is substantiated through rigorous experimental validation, achieving a root mean square (RMS) error between 5-10 cm, surpassing previous state-of-the-art systems significantly. The method's quantitative analysis was conducted over a combined dataset encompassing various conditions such as urban environments, GNSS-weak signals, and GNSS-denied areas. Specific outcomes involve comparison against existing methods, demonstrating the new system's superior performance in both regular and complex urban roads.

Implications and Future Directions

The implications of this research are noteworthy for the practical deployment of autonomous vehicles. Achieving high localization precision in diverse environments makes this system suitable for large-scale adoption in autonomous fleets, potentially enhancing navigation safety and reliability. The robustness of the multi-sensor fusion framework hints at its applicability to integrate additional sensors and accommodate lower-cost setups, broadening potential use cases.

Going forward, one possible direction of exploration involves integrating other sensor technologies or low-cost MEMS IMUs to further reduce the system's barrier to entry without compromising precision. Another area is the development of enhanced algorithms for real-time data fusion, optimizing computational efficiency for onboard vehicle systems. The authors also anticipate adapting the system towards ADAS or Level 3 autonomous driving capabilities.

In conclusion, the paper presents substantial progress in vehicle localization through multi-sensor fusion, achieving substantial precision and robustness necessary for diverse urban settings. This work not only contributes towards advancing autonomous vehicle technology but also provides a foundational framework that could be expanded and adapted to meet future demands in intelligent transportation systems.