A methodology for estimating the sideslip angle using a linear Kalman Filter with a model adaptation is presented. In this adaptation, neural networks are trained to estimate the model mismatch and standard deviations, which furthermore enable the calculation of the process and measurement covariance matrices. A particularly suitable mathematical formulation of the tire model is used for this purpose. The algorithm was trained and validated with measurements including different maneuvers and various road conditions. The results show that the algorithm is able to detect the sideslip angle on different road surfaces without knowledge of the current tire-road coefficient of friction. Under high excitations on ice, neural networks may occasionally fail, likely due to insufficient training data on ice. For this and similar situations, a fallback strategy is implemented and its effectiveness is demonstrated with measurements.

The sideslip angle is an important indicator of the stability and manoeuvrability of a vehicle. Its knowledge can improve vehicle dynamics controls and automated driving functions. An observer-based estimator for vehicle sideslip angle estimation is presented that only requires measurement data available in series production vehicles equipped with Electronic Stability Control. The estimator scheme uses two linear Kalman filters in a cascaded way. A phenomenological tyre model is proposed that is linear with regard to the states in the observer, but able to deal with tyre force nonlinearity as well as influences due to transient dynamics, road condition changes, and changes in vertical load. It is proven that the estimation error is bounded using the stability analysis. Results from vehicle tests show that the observer works accurately and robustly for different driving manoeuvres and on different road conditions. A parameter sensitivity study shows that the proposed observer strategy outperforms a state-of-the-art observer with regard to robustness to parameter changes.

Online tire model learning is of great importance in the application of vehicle dynamics related automated driving functions. Previous research mainly focuses on static tire model identification, for example, based on brush model. However, tire transient behavior, which is difficult to measure directly, plays a critical role on vehicle lateral dynamics, especially on safety-critical handling conditions. In this work, we propose a framework for online tire model iterative learning considering tire relaxation. In each steering scenario, we utilize the learning results of tire model from previous ones as initial condition. Subsequently, we implement singular value decomposition to detect whether there is enough excitation for model learning update with different simplified tire relaxation models. Meanwhile, based on these models, we also recursively calculate and compare the least squares cost function, such that the tire parameters can be robustly estimated and optimized. Furthermore, the estimated tire parameters are then fused with those from previous scenarios based on recursive average for better control application in the next steering maneuver. Experiments demonstrate the proposed online tire model iterative learning framework has similar performance with large-scale data-based offline fitting and can be applied for better predicting tire forces than those with purely recursive least squares methods.

Real-time trajectory planning in highly dynamic poses high challenges for the safe performance of highly automated vehicles, especially when considering the limited computational resources. In this paper, we propose a novel optimization-free trajectory planning approach for highly automated vehicles that integrates B & eacute;zier curve-based path planning with constrained backstepping-control-based velocity planning. Firstly, by leveraging the geometric properties of B & eacute;zier curves, we ensure several smooth and static-obstacle-collision-free paths. Subsequently, we implement backstepping control to plan the velocity considering states and control input constraints for selected paths. Finally, the planned trajectories are evaluated, and the optimal one is chosen based on predefined rules. This optimization-free property makes the proposed method update the trajectory at high frequencies, improving the ability to deal with highly dynamic environments. Simulation results based on Carsim demonstrate the effectiveness of the proposed method.

A limitation of simultaneous localization and mapping (SLAM) is the lack of consideration of dynamic objects in the environment, resulting in degradation of map quality and positioning accuracy. This article introduces a 2-D tightly-coupled LiDAR-Inertial Odometry and Mapping framework (DY-LIO) to remove dynamic objects for a better SLAM. The main innovations are based on the process of LiDAR point cloud and occupancy grid map: 1) with respect to LiDAR point cloud processing, based on deskewing and aligning the point cloud through factor-graph-optimized inertial measurement unit (IMU) preintegration, the alignment errors are utilized to identify dynamic objects. Besides, a different property of static and dynamic cloud points is introduced to reduce the misidentification error and 2) in terms of occupancy grid map processing, a novel probabilistic update method is presented to enhance the filterability of dynamic grid and accelerate the filtering process. Based on these novel strategies, DY-LIO can guarantee accurate localization and clean map estimation in dynamic environments. The experimental results on realistic high dynamic datasets demonstrate that DY-LIO greatly outperforms the current mainstream 2D-SLAM algorithm (Cartographer) in terms of the mapping effect and localization accuracy, with high efficiency in dynamic environments.

Dangerous driving behavior is a serious issue leading to harm drivers and further increase traffic burden. A You Only Look Once (YOLO) model is a commonly used fast detection model suitable for real-time dangerous driving behavior detection with poor detection performance. To address this problem, a lightweight object detection model called multiple fusion YOLO (MF-YOLO) model is proposed to show the superior capability in small target detection and compatibility with mobile chipsets. First, we design a novel backbone using convolution and vision transformer (ViT) multifusion blocks to fuse local and global context information. Second, a lightweight feature pyramid network (FPN) neck is developed to reduce model complexity and enhance feature extraction ability. Third, an attention mechanism is added to the neck for concentrating the YOLO model on relevant information during feature fusion. Finally, the activation function of fractional rectified linear unit (FReLU) equipped with spatial intersection over union (SIoU) loss function to improve model speed and accuracy. Experimental results from our self-built driving scenario dataset indicate that MF-YOLO achieved mean average precision (mAP) of 91.4%, surpassing YOLOv5n by 6.4%, and even outperforming the latest YOLOv8n by 2.3%.

The vehicle sideslip angle or lateral velocity is a measure both for driving stability and for occupant’s subjective perception of safety. With the introduction of vehicle dynamics control systems and automated driving functions, knowledge of this vehicle motion state is required for many control strategies. This article gives an overview on the state of the art on sideslip angle estimation. In contrast to other literature studies on this topic, it focuses on vehicle dynamics based algorithms. The following types of observers are discussed: Kalman Filter-type, recursive least squares (RLS), sliding mode observers (SMO) or nonlinear observers (NLO). Eventually, cascaded observers are used that first estimate some states, which then act as input to the sideslip angle estimator. Since the choice of an observer strategy always depends on the application, this article provides a brief insight into the work of selected research groups that have studied the topic. These examples will help to clarify the presence of many different approaches in the literature. A detailed discussion on vehicle and tire models is not included but referenced to other sources. Finally, this article provides recommendations for two main target groups: First, researchers and engineers that plan to design an algorithm for sideslip angle estimation using deterministic vehicle dynamics based approaches. Second, researchers and engineers planning to include an existing algorithm in an automated driving function that want to learn about advantages and limitations of these types of algorithms.

The tire-road friction coefficient (TRFC) is one of the most important parameters for intelligent distributed drive electric vehicles. The real-time performance of its estimation greatly affects motion control and trajectory planning. Traditional methods based on vehicle dynamics often only use current information for TRFC estimation. Due to measurement noises and insufficient driving excitation, convergent speed of estimation results cannot be set arbitrary fast. To this end, this paper proposes a TRFC estimator based on vehicle longitudinal dynamics which introduces previous information to improve the estimation convergent speed. Based on the previous wheel speed, slip rate as well as driving/braking torque information, this method designs a nonlinear adaptive observer that simultaneously estimates the historical tire longitudinal force and the current road friction coefficient in a receding horizon way. Through various simulations based on Carsim and Simulink, the results of the proposed estimation method is better than those from traditional method which only utilize current information in terms of estimation convergence speed.

Driver fatigue has long been recognized as a major cause of severe traffic accidents. However, achieving precise real-time fatigue detection for in-cabin drivers using low-cost devices remains a significant challenge. To tackle this issue, we propose a fatigue detection system for drivers based on temporal facial features. In the initial phase, we integrate a high-precision landmark model trained using a teacher-student distillation model with the You Only Look Once 5 face (YOLO5face) model to rapidly and accurately extract facial landmarks. Subsequently, a series of specific criteria are defined for classifying sequential parameters related to eye movement, mouth activity, and head posture. Extensive validation on both self-built dataset and four publicly available datasets demonstrates that both eye and head poses can be accurately detected, achieving an impressive accuracy rate of 97.1% in yawning detection. Furthermore, the system meets real-time detection requirements, operating at inference speed of 8-20 ms on a standard CPU. The source code is available at: https://github.com/Benpowder/Temporal-Facial-Features-Based-Fatigue-Detection-System.

Combining vehicle-dynamics-based methods (VDM) with camera-based methods (CBM) for a better road friction coefficient (RFC) estimation is a trend for safe automated driving. However, misclassification of road condition in CBM and reliable detection of driving excitation in VDM are not well considered, leading to poor RFC estimation and thus causing accidents in safety-critical scenarios. To overcome such problems, this work proposes a robust framework to estimate RFC and then applies it for safety-critical trajectory planning. Firstly, the RFC is estimated with a stable nonlinear estimator based on robust excitation detection with VDM. Then, RFC from VDM and CBM are fused considering camera mis-classification. The estimation of RFC is subsequently applied for safety-critical trajectory planning with two-stage model predictive control. Simulations based on CarSim demonstrate that the proposed framework can better guarantee planning safety than CBM and VDM combined method without considering camera mis-classification or reliable excitation detection.

To make a trajectory enough to track unmanned skid-steering vehicles, a planning method is proposed in this paper. The planner consists of velocity planning, curvature planning, and path planning. Subject to the constraints of jerk and acceleration, velocity planning has five possibilities. Then taking curvature constraints into account, curvature planning has three possibilities. And according to the velocity sequence and curvature sequence, the path information of the trajectory can be obtained by the kinematic model. To verify the presented method, a simulation scenario is designed. The test result shows that the proposed method could generate a sufficiently smooth trajectory with satisfying all the constraints, including acceleration, jerk, and curvature.

Simultaneous localization and mapping (SLAM) is a critical technology in the field of robotics. Over the past decades, numerous SLAM algorithms based on 2D LiDAR have been proposed. In general, these algorithms achieve good results in indoor environments. However, for geometrically degenerated environments such as long hallways, robust localization of robots remains a challenging problem. In this study, we focus on the challenges faced by LiDAR SLAM in such conditions. We propose 2D LiDAR SLAM algorithm Lmapping that employs an IMU-centric data-processing pipeline. In the front-end, a point cloud is directly registered to a probabilistic map; environment recognition is accomplished using a new method that relies on LiDAR measurements. And this method is more suitable for front-end matching based on grid maps. LiDAR odometry and IMU pre-integration are then integrated to build a local factor graph in the sliding window of the submap. When the environment is degraded, an Error-State Kalman Filter (ESKF) is added as a constraint to correct the IMU bias. In the back-end, through mutual matching within and between submaps, and loop detection, accumulated errors from the front-end are reduced. To improve flexibility for different sensor combinations, Lmapping supports multiple LiDAR inputs and facilitates initialization with a common six-axis IMU. Extensive experiments have shown that Lmapping greatly outperforms the current mainstream 2D-SLAM algorithm (Cartographer) in terms of the mapping effect and localization accuracy, with high efficiency in degraded environments.

This paper reports a two-stage damage detection method for effectively identifying multi-damage in composite plate structures. The first stage focuses on detecting damage locations using the wavelet transform to reveal the singularities from a modal shape of the composite plate. In the second stage, extreme learning machine (ELM) is employed to search the damage severities from the pre-calculated damage evaluation database (the relationship between the natural frequencies and severities). Using finite element method (FEM) and experimental modal analysis (EMA), the numerical simulations and experimental investigations are executed to verify the effectiveness of the combination method for simply supported composite plates structures with various damages. Results show that the proposed two-stage method is effectively to identify multiple damages in composite plate structures with reasonable precision might be extend to real-world applications.

Model predictive control (MPC) is widely used in automated driving due to its capability of dealing with multi-objectives and system constraints. To improve MPC's ability in tackling road friction uncertainty, contingency MPC (CMPC) is proposed by Alsterda [1]. However, current CMPC framework assumes the tire cornering stiffness is constant in the prediction horizon, which may cause large prediction error in the highly nonlinear tire region and thus may lead to a collision. In this paper, we propose a method to overcome this shortcoming by introducing varying cornering stiffness in the safety critical scenarios. Firstly, according to the upper and lower boundary of road condition as well as reference path, we deduce the steady state front and rear axle lateral force, based on which reference varying cornering stiffness is calculated. Secondly, CMPC is formulated by simultaneously considering tire cornering stiffness variation and road friction uncertainty in the prediction horizon. Finally, simulation results illustrate that our method performs better in safety critical situation than nominal CMPC does.

Automated and cooperative driving is one of the most promising but challenging tasks of automotive industry. For guaranteeing safe operation of automated driving, observing vehicle and environment states is of necessity, in which the tire-road friction coefficient (µ max ) is a crucial parameter. We give a solution to this problem by using the estimation framework proposed by the authors [1] for estimating the µ max by utilizing total aligning torque information and experimentally validate this observer. Firstly, we briefly introduce the proposed nonlinear adaptive observer. Then, a robust activation criteria is applied for reliable estimation. The estimation results from the proposed observer and Extended Kalman Filter (EKF) are subsequently compared under various µ max with experiments. The results show that 1) stability can be guaranteed with the proposed observer in various maneuvers while the EKF cannot. 2) The observer performs similar in terms of root mean square estimation error compared to EKF (when EKF is stable). Finally, detailed discussions are conducted to describe the potential limitations of the proposed method for µ max estimation.

In this paper, a real-time local path planner for obstacle avoidance of autonomous vehicles is proposed. Based on the tentacle algorithm, a novel tentacle curve that can ensure the continuity of the derivative of the path curvature is used to generate an alternate path set. After collision detection, we use the analytic hierarchy process (AHP) to calculate the weights to choose the best path. Finally, to verify the efficiency of the proposed planner, an obstacle avoidance case is designed in the simulation. The simulation results show that the proposed planner can obviously avoid the obstacle well and has better smoothness than the tentacle approach that uses the clothoid curve. The total time of local planning within a single period is less than 20 ms, which still has great real-time performance.

An adaptive singular value decomposition (SVD) method is proposed to detect damage based on the differential evolutionary algorithm for beam structures. B-spline wavelet on the interval finite element method is applied to model the damaged structures. The modal shape data obtained by modal analysis are used to construct attractor trajectory matrix, and SVD-based method is adopted to detect damage locations. In order to accommodate to different damage cases, differential evolutionary algorithm (DE) is used to adaptively optimize the parameters of SVD-based method. Both simulation and experiment researches are set up to verify the proposed method. The results indicate that the presented method is accurately effective to identify the damage for beam structures.

Damage to a compressor impeller can sometimes cause serious accidents, heavy casualties and property loss, etc. Therefore, it is necessary to conduct damage monitoring and identification for the compressor impeller. A damage identification method based on probabilistic neural networks (PNNs) with modal information fusion is proposed for a compressor impeller. The modal shape of the compressor impeller can be acquired by experimental modal analysis. Combining waveform capacity dimension, a singular value decomposition is applied to extract damage feature information from the system modal shape. The two damage indicators are fused by a multi-dimensional feature vector. Finally, a PNN model is constructed and used to identify structural damage. The experimental results indicate that the proposed method is effective in detecting damage to the compressor impeller.

Lithium-ion battery state of charge (SOC) is an important parameter reflecting the characteristics of battery and power system, which is closely related to battery safety, life and efficiency. However, it is normally difficult to guarantee stability of SOC estimation error with Extended Kalman Filter (EKF) based on second-order RC equivalent model. This paper aims at solving the stability problem of SOC estimation, by introducing eXogenous Kalman filter (XKF). This XKF includes two steps. Firstly, SOC is observed by a stable observer with relatively poor quality, since model and parameter uncertainty is not considered. Secondly, the poor SOC results are fed into EKF for obtaining a stable and accurate SOC estimation by considering model and parameter mismatch. Experiments demonstrate that results of SOC estimation from XKF are accurate and can guarantee stability while EKF cannot.

The tire-road friction coefficient (μmax) is an important input for vehicle dynamics control system and automated driving modules. However, reliable and accurate measurement of this parameter is difficult and costly in mass-produced vehicles and thus estimation is necessary. In this research, an innovative optimization based framework to estimate μmax is proposed. The observation problem is formulated as a non-convex optimization. A novelty of the framework is that the μmax can be accurately estimated in real time together with side slip angle as a by-product without requiring a good initial guess for the non-convex optimization. A key observation is that the time derivative of μmax and side slip angle can be assumed as zero and computed based on measurement, respectively. This allows the observed variables to be updated at a relatively low frequency w.r.t. the solution of the optimization problem. During the interval between each two neighbouring updating time, the observer estimates the μmax and side slip angle by integrating sensor information based on the last update. To find the global optima approximately, a grid search method is implemented for solving non-convex optimization. The estimation results from the proposed observer and a linearization based observer (lbo) are finally compared under various tire-road conditions with simulations and experiments. The results showed that 1) the proposed observer can always guarantee stability in a wide range of vehicle operations while lbo cannot. 2) w.r.t. root mean square of estimation error, the proposed observer performs overall better than lbo in μmax estimation.