Accurate and robust localization using multi-modal sensors is crucial for autonomous driving applications. Although wheel encoder measurements can provide additional velocity information for visual-inertial odometry (VIO), the existing visual-inertial-wheel odometry (VIWO) still cannot avoid long-term drift caused by the low-precision attitude acquired by the gyroscope of a low-cost inertial measurement unit (IMU), especially in visually restricted scenes where the visual information cannot accurately correct for the IMU bias. In this work, leveraging the powerful data processing capability of deep learning, we propose a novel tightly coupled monocular visual-inertial-wheel odometry with neural gyroscope calibration (NGC) to obtain accurate, robust, and long-term localization for autonomous vehicles. First, to cure the drift of the gyroscope, we design a robust neural gyroscope calibration network for low-cost IMU gyroscope measurements (called NGC-Net). Following a carefully deduced mathematical calibration model, NGC-Net leverages the temporal convolutional network to extract different scale features from raw IMU measurements in the past and regress the gyroscope corrections to output the de-noised gyroscope. A series of experiments on public datasets show that our NGC-Net has better performance on gyroscope de-noising than learning methods and competes with state-of-the-art VIO methods. Moreover, based on the more accurate de-noised gyroscope, an effective strategy for combining the advantages of VIWO and NGC-Net outputs is proposed in a tightly coupled framework, which significantly improves the accuracy of the state-of-the-art VIO/VIWO methods. In long-term and large-scale urban environments, our RNGC-VIWO tracking system performs robustly, and experimental results demonstrate the superiority of our method in terms of robustness and accuracy.

Currently, most visual-based simultaneous localization and mapping (VSLAM) algorithms assume that the environment is largely static, while the scenes surrounding autonomous vehicles are usually highly dynamic. Therefore, this assumption greatly limits the application of VSLAM in autonomous driving. In this work, we propose a novel tightly-coupled monocular visual-inertial-wheel odometry framework, called Dyna-VIWO, which is robust against high-dynamic urban environments for autonomous vehicles. In this framework, we first propose a robust adaptive method for adjusting the residual weights of potential dynamic features, by jointly considering semantic constraints and pose prior estimated by the denoised IMU-Odometer preintegration. Then, a new system abnormality detection algorithm and state update strategy are proposed to ensure the smoothness and accuracy of pose estimation in highly dynamic scenes. In particular, the proportion of dynamic points participated in the optimizer serves as one of the main indicators for system anomaly judgment, and we develop a new dynamic point recognition algorithm based on semantic prior and geometric constraints between consecutive frames. Finally, we evaluate our method on a public dataset containing a large number of dynamic objects. The results show that our proposed method can significantly improve the robustness and accuracy of the system compared to the state-of-the-art VIO/VIWO methods in highly dynamic environments.

Lane detection is a fundamental task in computer vision with broad applications in autonomous driving. Recent advances have focused on line-anchor-based representations for their efficiency and high accuracy. However, most existing methods adopt anchors that are continuous in point sampling yet discrete in width and fixed in direction. This leads to geometric misalignments with real-world lane lines, which exhibit continuous width representation and smoothly varying directions. These misalignments cause Symmetric Point Ambiguity and Magnified Localization Errors, ultimately degrading both evaluation reliability and detection accuracy. To address these issues, we propose Angle-Aware Rectangle Anchors (ARA), a novel representation with continuous width and adaptive directional alignment, that effectively captures the geometric variation of lane lines. In addition, we propose the Three-phase Angle-Thresholded Line-Area Transition (TALAT) Loss, which dynamically switches between loss formulations based on angular thresholds and overlap quality, enabling a smooth transition from coarse to fine supervision. Extensive experiments on multiple standard benchmarks (TuSimple, CULane, CurveLanes, and LLAMAS) demonstrate that our approach achieves competitive or state-of-the-art (SOTA) performance. Notably, ARA exhibits superior robustness in challenging scenarios, while maintaining real-time inference speed. Our code and models are publicly available at https://github.com/changehome717/ARA-Lane-Detection.