HybVIO: Pushing the Limits of Real-time Visual-inertial Odometry

Abstract: We present HybVIO, a novel hybrid approach for combining filtering-based visual-inertial odometry (VIO) with optimization-based SLAM. The core of our method is highly robust, independent VIO with improved IMU bias modeling, outlier rejection, stationarity detection, and feature track selection, which is adjustable to run on embedded hardware. Long-term consistency is achieved with a loosely-coupled SLAM module. In academic benchmarks, our solution yields excellent performance in all categories, especially in the real-time use case, where we outperform the current state-of-the-art. We also demonstrate the feasibility of VIO for vehicular tracking on consumer-grade hardware using a custom dataset, and show good performance in comparison to current commercial VISLAM alternatives. An open-source implementation of the HybVIO method is available at https://github.com/SpectacularAI/HybVIO

Overview of the HybVIO: Hybrid Visual-Inertial Odometry and SLAM Approaches

In the paper, the authors present a novel methodology known as HybVIO, which integrates filtering-based visual-inertial odometry (VIO) with optimization-based simultaneous localization and mapping (SLAM). The need for such hybrid approaches arises due to limitations in solely relying on VIO or SLAM systems, where VIO is particularly efficient for short-term motion tracking and SLAM excels in providing long-term consistency through map creation and loop closure.

Methodological Advancements

The authors make a series of methodological contributions that enhance both the efficacy and efficiency of visual-inertial navigation and mapping systems:

1. Integration of Probabilistic Framework: The proposed system extends the probabilistic inertial-visual odometry (PIVO) framework to not only support monocular but also stereo camera data, thus broadening its application scope.
2. Refined IMU Bias Modeling: The paper introduces Ornstein-Uhlenbeck processes for more accurate IMU bias modeling, presenting a statistically grounded approach to dealing with bias drift, which is especially critical in applications demanding long-term precision.
3. Enhanced Feature and Outlier Management: New methodologies for outlier detection and feature track selection are introduced, improving robustness. These mechanisms are based within the probabilistic framework, allowing for a more informed decision-making process under uncertainty.
4. Loosely Coupled SLAM Module: By combining filtering-based VIO with optimization-based SLAM in a loosely-coupled fashion, the authors attain a balance between real-time performance and map accuracy. This hybrid method manages to yield superior results compared to more traditional approaches.

Performance and Results

The HybVIO system demonstrated significant advantages in academic benchmarks, surpassing other state-of-the-art approaches. Notably, in the EuRoC MAV dataset, it outperformed competitors in real-time stereo tracking. It exhibited competitive performance in both monocular and stereo settings across both online and post-processed categories, emphasizing its versatility and robustness. The system’s capability was further validated through tests on consumer-grade platforms, underscoring its applicability in vehicular contexts using available commercial hardware.

Practical Implications

The integration of probabilistic methods addresses the challenges of real-world deployments where systems are vulnerable to sensor inaccuracies and environmental dynamics. The HybVIO system, with its reduced computational load and ability to function on embedded platforms such as Raspberry Pi, presents a pragmatic solution that blends computational efficiency with algorithmic complexity.

Future Developments

Acknowledging the constraints and simplifications within the current implementation, future developments could explore the integration of novel initialization techniques, the utilization of first-estimate Jacobians, and pose estimation through optimization on SO(3) manifolds. Furthermore, the expansion of loop closure techniques and extended SLAM capabilities might enhance the method's applicability to more challenging environments beyond the scope of existing datasets.

Conclusion

The paper illustrates a successful endeavor in merging the distinct strengths of filtering and optimization approaches to visual-inertial odometry and SLAM. The contributions provide a foundational leap towards more adaptive, flexible, and competent autonomous navigation systems, promising potential enhancements and applications in diverse use cases ranging from augmented reality to autonomous vehicle navigation.