Susu Fang

Accurate and robust moving object detection and tracking are critical measurement tasks for autonomous vehicles operating in complex dynamic scenes. Cooperative perception, leveraging multi-vehicle information sharing, offers a way to overcome inherent limitations of single-vehicle measurement systems. This paper presents a cooperative measurement solution based on LiDAR point clouds and IMU, comprising a cooperative object detection module using temporally fused point clouds and an IMM-SCIF-JPDA based multi-object tracker. In the detection module, temporal point cloud information is integrated via IMU mechanization, and an adaptive feature-level fusion model is applied to improve detection precision while maintaining robustness under occlusion and sparse point cloud conditions. For the tracking module, an interactive multi-model (IMM) filter combined with the split covariance intersection filter (Split CIF) is developed to handle correlation problems and measurement uncertainty arising from unknown object motion patterns. This module is further integrated with joint probability data association (JPDA) to enhance tracking stability and accuracy. Comparative experimental results are provided to quantify the performance of the proposed cooperative perception solution, with detailed analysis of its improvements in both accuracy and robustness.

Simultaneous localization and mapping (SLAM) is a fundamental problem for mobile robots. In recent years, solid-state LiDAR have garnered growing attention owing to their enhanced reliability, cost-effectiveness and lightweight. However, in comparison to traditional rotating LiDAR, employing solid-state LiDAR for SLAM is challenging due to their small field of view (FoV) in the lateral direction. In unstructured environments that lack distinct features and geometric structures, small FoV introduces error in pose estimation, particularly in ill-conditioned directions, leading to LiDAR degeneration. Additionally, the small FoV leads to a higher proportion of dynamic points in the LiDAR frame, which also contributes to error in point cloud registration. To address these challenges, we propose a LiDAR-inertial-visual SLAM method for solid-state LiDAR, which overcomes limitations of small FoV. This method consists of a LiDAR-inertial odometry to provide geometric constraint and a visual-inertial odometry to provide visual constraint. To resist the LiDAR degeneration, a novel residual construction method is proposed, which enhances the dimensionality of residual in visual-inertial odometry. Furthermore, we propose a keyframe-based static point registration method to effectively decrease incorrect constraints from dynamic points in LiDAR-inertial odometry. Various comparative experiments demonstrate the performance and advantages of the proposed LiDAR-inertial-visual SLAM method in terms of robustness and accuracy.

Simultaneous localization and mapping (SLAM) and moving object detection and tracking (MODT) are two fundamental problems for autonomous driving systems. Multi-vehicle cooperative SLAM and cooperative object perception, which take advantage of multi-vehicle information sharing, can overcome inherent limitations of single vehicle perception, such as view occlusion. Solutions to SLAM and MODT usually rely on certain assumptions, such as the static environment assumption for SLAM and the accurate ego-vehicle pose assumption for MODT. However, it is difficult or even impossible to have these assumptions hold in complex dynamic environments. We propose a LiDAR-based coupled cooperative simultaneous SLAM and MODT (C-SLAMMODT) strategy, which not only handles the SLAM and tracking problems in dynamic environments but also overcomes limitations of single vehicle perception. The proposed C-SLAMMODT outperforms both cooperative SLAM and cooperative MODT. This method includes a cooperative SLAM module that can augment ego-vehicle pose estimation by shared information from neighbouring vehicles, a cooperative MODT module that applies a state-of-the-art adaptive feature-level fusion model to fuse multi-vehicle data, improving detection precision and overcoming the limitations of perception in occlusion situations. Furthermore, a unified factor graph optimization integrates the information obtained from ego-vehicle states, neighbor-vehicle shared data, and dynamic object states to augment pose estimation and realize object tracking. Various comparative experiments demonstrate the performance and advantages of the proposed C-SLAMMODT solution in terms of accuracy and robustness.

High-precision localization in an unknown environment is the fundamental requirement of autonomous vehicle for safe advanced driving. Cooperative localization has obvious advantages in precision, fault tolerance, and flexibility compared with single-vehicle localization, which estimates the vehicle pose via fusing the multiple sources data from sensors and extra shared neighbor vehicle information available when multiple vehicles operate simultaneously. For cooperative localization, there are more than one type of error data sources that degenerate the state estimation need to be considered, such as inter-estimation correlation and innovation, observation outliers, which may exist simultaneously. Previous works usually evaluate performance under one of the error types at one time, or they do not perform well under some extreme situations where both of the above-mentioned correlation and outliers exist. This paper presents an accurate and robust iterated split covariance intersection filter (Iterated Split CIF) based cooperative localization strategy with a decentralized framework, which can ensure the performance when data sources of various error types exist at the same time. In addition, we adopt an effective point cloud registration method to obtain the cooperative relative pose estimation using mutually shared information from neighbor vehicles. A CARLA simulator based comparative study demonstrates the potential and advantage of the proposed multi-vehicle cooperative localization using Iterated Split CIF in terms of accuracy, robustness and efficiency.

In autonomous vehicle navigation, it usually involves a basic process of estimating the vehicle pose at one instant relative to the vehicle pose at a previous instant, and we refer to this process as temporal relative localization (TRL). Accurate and reliable TRL is valuable for vehicle localization and mapping which plays a fundamental role in autonomous vehicle navigation. A common way for realizing TRL is dead reckoning based on motion sensor such as inertial navigation system (INS). To augment the performance of INS based TRL, a nine-degree-of-freedom (9-DOF) vehicle dynamics model (VDM) is proposed as the system model for the filtering process of TRL; it can effectively characterize vehicle pitch and roll motion, especially in slope road scenarios. For better fault tolerance, a distributed fusion framework which consists of a master filter and two local filters for multi-source data fusion is introduced. The split covariance intersection filter (Split CIF) is applied to handle unknown temporal and spatial correlation that inevitably exists in such distributed fusion mechanism. A comparative experimental study demonstrates the advantage of the proposed method in terms of accuracy and robustness.

Cooperative localization is still an open problem for autonomous vehicle driving tasks. Generally, there are two essential issues to be handled for multi-vehicle cooperative localization: one is how to use extra information shared by neighbour vehicles to augment ego-vehicle localization; the other is how to handle data sources of various error types used for state estimation, such as nonlinearity, dynamic noises, and correlation which may exist simultaneously. Previous works usually evaluate performance under one of the issues at one time, or they do not perform well under some extreme situations with both of the two types of problems. This paper presents an accurate and robust cooperative localization system with a decentralized framework, which not only takes advantages of inter vehicle relative pose estimation via shared information, but also ensures the performance under data sources of various error types simultaneously. We adopt a state-of-the-art point cloud registration method to obtain the cooperative relative pose estimation using shared information. In addition, we use the adaptive cubature split covariance intersection filter (ACSCIF) to estimate the vehicle state. This algorithm can approximate nonlinearity and deal with dynamic measurement noise by a third-order term using cubature transformation and innovation-based adaptive estimation (IAE) theory, also considering correlations among estimated sources inevitably exists in the decentralized mechanism. Finally, a comparative study based on the CARLA simulation platform demonstrates the potential and advantage of the proposed multi-vehicle cooperative localization system using ACSCIF in terms of accuracy, robustness and efficiency.

To improve the accuracy of attitude and heading reference systems for moving vehicles, an effective orientation estimation method is proposed. The method uses an odometer, a low-cost magnetic, angular rate, and gravity sensor. This study addresses the problems of non-orthogonal error, carrier magnetic field interference and calibration to obtain accurate, long-term, stable magnetic field strength. A neural network fusion 12-parameter ellipse fitting method is proposed to eliminate the soft magnetic field and hard magnetic field interference. The interference to the accelerometer from linear acceleration is eliminated by using an odometer and a gyroscope, and the high-frequency noise from the accelerometer is eliminated by using a low-pass filter. An improved method to evaluate vehicle attitude is proposed and utilized to compensate for filtered accelerometer measurement when the vehicle is moving at a uniform, accelerate and steering state. The proposed method uses an effective adaptive Kalman filter based on the error state model to reduce dynamic perturbations. Filter gain is adaptively tuned under different moving modes by adjusting the noise matrix. The effectiveness of the algorithm is verified by experiments and simulations in multiple operating conditions.

When a low-cost micro-electro-mechanical system inertial measurement unit is used for a vehicle navigation system, errors will quickly accumulate because of the large micro-electro-mechanical system sensor measurement noise. To solve this problem, an automotive sensor–aided low-cost inertial navigation system is proposed in this article. The error-state model of the strapdown inertial navigation system has been derived, and the measurements from the wheel speed sensor and steer angle sensor are used as the new observation vector. Then, the micro-electro-mechanical system inertial measurement unit/wheel speed sensor/steer angle sensor–integrated system is established based on adaptive Kalman filtering. The experimental results show that the positioning error of micro-electro-mechanical system inertial measurement unit/wheel speed sensor/steer angle sensor is 94.67%, 98.88%, and 97.88% less than the values using pure strapdown inertial navigation system in the east, north, and down directions, respectively. The yaw angle error is reduced to less than 1°, and the vehicle velocity estimation of micro-electro-mechanical system inertial measurement unit/wheel speed sensor/steer angle sensor–integrated navigation system is closer to the reference value. These results show the precision of the integrated navigation solution.

A constrained low‐cost SINS/OD filter aided with magnetometer is proposed in this paper. The filter is designed to provide a land vehicle navigation solution by fusing the measurements of the microelectromechanical systems based inertial measurement unit (MEMS IMU), the magnetometer (MAG), and the velocity measurement from odometer (OD). First, accelerometer and magnetometer integrated algorithm is studied to stabilize the attitude angle. Next, a SINS/OD/MAG integrated navigation system is designed and simulated, using an adaptive Kalman filter (AKF). It is shown that the accuracy of the integrated navigation system will be implemented to some extent. The field‐test shows that the azimuth misalignment angle will diminish to less than 1°. Finally, an outliers detection algorithm is studied to estimate the velocity measurement bias of the odometer. The experimental results show the enhancement in restraining observation outliers that improves the precision of the integrated navigation system.