A General Optimization-based Framework for Global Pose Estimation with Multiple Sensors (1901.03642v1)

Abstract: Accurate state estimation is a fundamental problem for autonomous robots. To achieve locally accurate and globally drift-free state estimation, multiple sensors with complementary properties are usually fused together. Local sensors (camera, IMU, LiDAR, etc) provide precise pose within a small region, while global sensors (GPS, magnetometer, barometer, etc) supply noisy but globally drift-free localization in a large-scale environment. In this paper, we propose a sensor fusion framework to fuse local states with global sensors, which achieves locally accurate and globally drift-free pose estimation. Local estimations, produced by existing VO/VIO approaches, are fused with global sensors in a pose graph optimization. Within the graph optimization, local estimations are aligned into a global coordinate. Meanwhile, the accumulated drifts are eliminated. We evaluate the performance of our system on public datasets and with real-world experiments. Results are compared against other state-of-the-art algorithms. We highlight that our system is a general framework, which can easily fuse various global sensors in a unified pose graph optimization. Our implementations are open source\footnote{https://github.com/HKUST-Aerial-Robotics/VINS-Fusion}.

• The paper introduces an optimization-based framework that fuses local VO/VIO and global sensor data for accurate global pose estimation.
• It employs pose graph optimization using the Ceres Solver and Levenberg-Marquardt algorithm to effectively reduce drift in state estimation.
• Experimental results on the KITTI dataset and outdoor trials demonstrate improved localization performance over state-of-the-art methods.

Overview of Optimization Framework for Multi-Sensor Global Pose Estimation

The paper authored by Tong Qin, Shaozu Cao, Jie Pan, and Shaojie Shen presents an optimization-based framework designed for integrating multiple sensor inputs to achieve precise global pose estimation for autonomous robots. Recognizing the criticality of accurate state estimation in autonomous applications, the authors propose a novel sensor fusion approach that combines both local and global sensor outputs to mitigate the limitations associated with each type.

Problem Statement and Motivation

Autonomous systems necessitate precise localization to fulfill tasks such as navigation and control effectively. Local sensors like cameras, IMUs, and LiDARs provide high-precision pose estimates within restricted regions. However, these systems suffer from accumulated drift over time in the absence of a global frame reference. Global sensors such as GPS and magnetometers, conversely, offer drift-free measurements but are hindered by noise and low-frequency updates, limiting their capability to support real-time, high-precision localization independently. The paper addresses the challenge of harmonizing these disparate systems to deliver locally precise and globally stable pose estimation.

Framework Construction

The authors propose a comprehensive sensor fusion framework that deploys pose graph optimization to synergize local estimations from VO/VIO with global sensor readings, such as GPS. The novel aspect of this framework is its flexibility and adaptability, allowing the fusion of a wide variety of sensors into a unified optimization schema. The local estimations are transformed into a global coordinate frame, minimizing accumulated drift.

Methodology

1. Local Pose Estimation: This involves leveraging existing VO/VIO techniques, where the emphasis is on maintaining high accuracy within small regions. The framework uses positional and orientation data from IMUs and visual sensors.
2. Global Pose Graph: It organizes system states into nodes representing poses, with edges denoting local and global sensor-derived pose constraints. This structure is effective in aligning local estimation within a broader global framework to reduce drift.
3. Sensor Factors Integration:
   • Local Factors: Emphasizing the relative poses derived from VO/VIO.
   • Global Factors: Transforming GPS, magnetometer, and barometer outputs into optimization constraints reflecting global measurements.

The optimization challenge is approached using an iterative method via Ceres Solver, taking advantage of efficient computational techniques like Levenberg-Marquardt.

Experimental Validation

The framework is validated through experiments utilizing both the KITTI dataset, known for its comprehensive collection of labeled automotive data, and real-world outdoor experiments. Comparisons with state-of-the-art methods, specifically ORB-SLAM and MSF, demonstrated the proposed framework’s competitive edge in reducing translation drift across extensive test sequences.

Implications and Future Directives

This framework advances the field by providing a robust system that accommodates multiple sensor types, easing the integration of new sensors into existing systems. Its potential impact spans various autonomous applications, such as vehicles and drones, requiring precise long-term localization.

Future work could further expand the scope of sensor inclusion and optimize processing efficiency for larger, more complex environments, potentially making the system scalable for industrial and urban outdoor settings.

In summary, this paper contributes a versatile and efficient methodology for global pose estimation, showcasing significant improvements in localization performance, particularly in drift reduction, by integrating multiple sensory inputs into a single coherent optimization framework.