Ground-Fusion: A Low-cost Ground SLAM System Robust to Corner Cases (2402.14308v1)

Abstract: We introduce Ground-Fusion, a low-cost sensor fusion simultaneous localization and mapping (SLAM) system for ground vehicles. Our system features efficient initialization, effective sensor anomaly detection and handling, real-time dense color mapping, and robust localization in diverse environments. We tightly integrate RGB-D images, inertial measurements, wheel odometer and GNSS signals within a factor graph to achieve accurate and reliable localization both indoors and outdoors. To ensure successful initialization, we propose an efficient strategy that comprises three different methods: stationary, visual, and dynamic, tailored to handle diverse cases. Furthermore, we develop mechanisms to detect sensor anomalies and degradation, handling them adeptly to maintain system accuracy. Our experimental results on both public and self-collected datasets demonstrate that Ground-Fusion outperforms existing low-cost SLAM systems in corner cases. We release the code and datasets at https://github.com/SJTU-ViSYS/Ground-Fusion.

• The paper presents an adaptable SLAM system that fuses RGB-D, IMU, odometry, and GNSS data for robust ground localization.
• It employs adaptive initialization and real-time dense color mapping to effectively mitigate sensor faults like wheel slip and GNSS degradation.
• Experimental results demonstrate superior performance in challenging corner cases, enhancing reliability during transitions between indoor and outdoor settings.

Efficient and Robust Ground-Fusion SLAM System Adapting to Diverse Environments and Anomalies

Introduction

Simultaneous Localization and Mapping (SLAM) is a core technology enabling ground robots to navigate effectively across diverse settings, from indoor corridors to outdoor terrains. Recent advancements have largely revolved around the integration of various sensing technologies to improve accuracy and reliability. The introduction of Ground-Fusion, a low-cost SLAM system, has made significant strides in this direction. By tightly coupling data from RGB-D cameras, inertial measurement units (IMUs), wheel odometers, and Global Navigation Satellite Systems (GNSS), Ground-Fusion presents an adaptable and precise solution capable of handling a variety of challenging scenarios and corner cases.

System Overview

Ground-Fusion leverages an innovative combination of sensing modalities within a factor graph optimization framework, focusing on efficient system initialization and robust sensor fusion. Key aspects include:

• Adaptive Initialization: Employing stationary, visual, and dynamic strategies ensures the system can initialize accurately across varied motion states and environmental conditions.
• Anomaly Detection and Handling: Sophisticated mechanisms are in place to identify and mitigate sensor faults or anomalies arising from visual obstacles, wheel slippage, or GNSS signal degradation.
• Real-time Dense Color Mapping: By incorporating dense color mapping, Ground-Fusion aids in better navigation and offers a clearer understanding of the robot's surroundings.

System Evaluation

Considerable experiments validate the effectiveness and robustness of Ground-Fusion. Compared to existing low-cost SLAM systems, Ground-Fusion demonstrates superior performance, particularly in corner cases. Scenarios including transitions from outdoor to indoor environments, areas with dense tree cover affecting GNSS reliability, and settings prone to wheel anomalies underpin its robustness. Furthermore, the system's ability to efficiently handle low-speed movements and zero-velocity updates accentuates its adaptability.

Methodological Contributions

Ground-Fusion's methodology stands out in several ways:

• Its approach to initialization is divergent, leveraging an adaptive mechanism that responds dynamically to the vehicle's motion state.
• The system's anomaly detection and handling offer a nuanced way of maintaining accuracy even when individual sensor data might be compromised.
• Through the strategic fusion of multi-sensor data, Ground-Fusion maintains a high level of accuracy in localization, effectively minimizing the trade-offs associated with using low-cost sensors.

Implications and Future Directions

The development of Ground-Fusion has practical implications for the deployment of ground robots across various industries, including logistics and manufacturing. Its ability to navigate reliably in both indoor and outdoor settings, coupled with its resilience to sensor anomalies, makes it particularly valuable where cost and operational efficiency are paramount.

The foundation laid by Ground-Fusion paves the way for future research in a few key areas. Exploring advanced sensor fusion algorithms that can further minimize the influence of environmental anomalies on localization accuracy remains a critical path. Additionally, enhancing the system's capability to handle dynamic and unpredictable environments through machine learning techniques could offer new avenues for improvement.

Conclusion

Ground-Fusion represents a significant step forward in the development of reliable, low-cost SLAM systems for ground vehicles. By intelligently integrating multiple sensors and emphasizing robustness in the face of anomalies, it establishes a new benchmark for performance in challenging conditions. As SLAM technology continues to evolve, Ground-Fusion lays the groundwork for future innovations aimed at creating even more versatile and resilient navigation systems for autonomous robots.