A two-stage-training support vector machine approach to predicting unintentional vehicle lane departure

Advanced driver assistance systems, such as unintentional lane departure warning systems, have recently drawn much attention and efforts. In this study, we explored utilizing the nonlinear binary support vector machine (SVM) technique to predict unintentional lane departure, which is innovative, as the SVM methodology has not previously been attempted for this purpose in the literature. Furthermore, we developed a two-stage training scheme to improve SVM's prediction performance in terms of minimization of the number of false positive prediction errors. Experiment data generated by VIRTTEX, a hydraulically powered, 6-degrees-of-freedom moving base driving simulator at Ford Motor Company, were used. All the vehicle variables were sampled at 50 Hz and there were 16 drowsy drivers (about 3 hours of driving per subject) and six control drivers (approximately 20 minutes f driving each). In total, 3,508 unintentional lane departures occurred for the drowsy drivers and 23 for the control drivers. Our study involving these 22 drivers with a total of more than 7.5 million prediction decisions demonstrates that (a) excellent SVM prediction performance, measured by numbers of false positives (i.e., falsely predicted lane departures) and false negatives (i.e., lane departures failed to be predicted), was achieved when the prediction horizon was 0.6 seconds or less, (b) lateral position and lateral velocity worked the best as SVM input variables among the nine variable sets that we explored, and (c) the radial basis function performed the best as the SVM kernel function.     

Road curvature estimation for vehicle lane departure detection using a robust Takagi–Sugeno fuzzy observer

In this paper, a lane departure detection method is studied and evaluated via a professional vehicle dynamics software. Based on a robust fuzzy observer designed with unmeasurable premise variables with unknown inputs, the road curvature is estimated and compared with the vehicle trajectory curvature. The difference between the two curvatures is used by the proposed algorithm as the first driving risk indicator. To reduce false alarms and take into account the driver corrections, a second driving risk indicator is considered, which is based on the steering dynamics, and it gives the time to the lane keeping. The used nonlinear model deduced from the vehicle lateral dynamics and a vision system is represented by an uncertain Takagi–Sugeno fuzzy model. Taking into account the unmeasured variables, an unknown input fuzzy observer is then proposed. Synthesis conditions of the proposed fuzzy observer are formulated in terms of linear matrix inequalities using Lyapunov method. The proposed approach is evaluated under different driving scenarios using a software simulator. Simulation results show good efficiency of the proposed method.     

A new shared control for lane keeping and road departure prevention

In recent years, the driver's active assistances have become important features in commercialised vehicles. In this paper, we present one of these features which consists of an advanced driver assistance system for lane keeping. A thorough analysis of its performance and stability with respect to variations in driver behaviour will be given. Firstly, the lateral control model based on visual preview is established and the kinematics model based on visual preview, including speed and other factors, is used to calculate the lateral error and direction error. Secondly, and according to the characteristics of the lateral control, an efficient strategy of intelligent electric vehicle lateral mode is proposed. The integration of the vehicle current lateral error and direction error is chosen as the parameter of the sliding mode switching function to design the sliding surface. The control variables are adjusted according to the fuzzy control rules to ensure that they meet the existence and reaching condition. A new fuzzy logic-based switching strategy with an efficient control law is also proposed to ensure a level of continuous and variable sharing according to the state of the driver and the vehicle positioning on the roadway. The proposed control law acts either at the centre of the lane, as a lane keeping assistance system to reduce the driver's workload for long trips, or as a lane departure avoidance system that intervenes for unintended lane departures. Simulation results are included in this paper to explain this concept.     

Vehicle dynamics and road curvature estimation for lane departure warning system using robust fuzzy observers: experimental validation

This paper describes a new approach to estimate vehicle dynamics and the road curvature in order to detect vehicle lane departures. This method has been evaluated through an experimental set-up using a real test vehicle equipped with the RT2500 inertial measurement unit. Based on a robust unknown input fuzzy observer, the road curvature is estimated and compared to the vehicle trajectory curvature. The difference between the two curvatures is used by the proposed lane departure detection algorithm as the first driving risk indicator. To reduce false alarms and take into account driver corrections, a second driving risk indicator based on the steering dynamics is considered. The vehicle nonlinear model is deduced from the vehicle lateral dynamics and road geometry and then represented by an uncertain Takagi–Sugeno fuzzy model. Taking into account the unmeasured variables, an unknown input fuzzy observer is proposed. Synthesis conditions of the proposed fuzzy observer are formulated in terms of linear matrix inequalities using the Lyapunov method.     

智能安全车辆技术前沿跟踪

通过对智能安全车辆技术的起源和发展介绍。并针对有应用前景的智能安全车辆技术如：车道偏离预警系统、主动纵向避撞系统和自适应巡航控制系统,逐一进行了深入分析;给出了系统功能的定义和实现原理,指出了关键核心技术和实现难点所在。为我国开展实施智能安全车辆技术提出了建议和技术路线。     

A receding horizon sliding control approach for electric powertrains with backlash and flexible half-shafts

Active control of electric powertrains is challenging, due to the fact that backlash and structural flexibility in transmission components can cause severe performance degradation or even instability of the control system. Furthermore, high impact forces in transmissions reduce driving comfort and possibly lead to damage of the mechanical elements in contact. In this paper, a nonlinear electric powertrain is modelled as a piecewise affine (PWA) system. The novel receding horizon sliding control (RHSC) idea is extended to constrained PWA systems and utilised to systematically address the active control problem for electric powertrains. Simulations are conducted in Matlab/Simulink in conjunction with the high fidelity Carsim software. RHSC shows superior jerk suppression and target wheel speed tracking performance as well as reduced computational cost over classical model predictive control (MPC). This indicates the newly proposed RHSC is an effective method to address the active control problem for electric powertrains.     

基于分段学习模型的自动驾驶行为决策算法研究

在具有车道线的特定自动驾驶场景中,针对目前端到端的行为决策算法直接输入原始图像进行决策导致的网络模型迁移性差、预测精度欠佳、泛化能力不足等问题,提出一种基于分段学习模型的车辆自动驾驶行为决策算法。首先,基于GoogLeNet建立一种端到端的车道线检测网络模型,并引入车道中心线作为决策的重要线索提高算法的迁移能力,同时利用YOLOv3目标检测模型对本车道内前方最近障碍物进行位置检测;而后,经几何测量模型将两者检测结果转换成环境状态信息向量为决策做支撑;最后,构建基于长短期记忆(LSTM)网络的驾驶行为决策模型,根据编码的历史状态信息刻画出动态环境中车辆的运动模式,并结合当前时刻的状态推理得到驾驶行为参量。使用建立的真实驾驶场景数据集对模型分别进行训练、验证与测试,离线测试结果显示车道线检测模型的检测位置误差小于1.3%,车道内前方障碍物检测模型的检测精度达98%以上,驾驶行为决策网络模型表征预测优度的决定系数 大于0.7。为进一步验证算法的有效性,搭建了Simulink/PreScan联合仿真平台,多种工况下的仿真验证试验中多个评价指标均达到工程精度要求,实车测试的试验结果也表明该算法可实现复杂驾驶场景下平稳、准确无偏航的预测效果并满足实时性要求,且与传统端到端模式的算法相比,具有更好的迁移性和泛化能力。     

Optimal control of brakes and steering for autonomous collision avoidance using modified Hamiltonian algorithm

ABSTRACT

This paper considers the problem of collision avoidance for road vehicles, operating at the limits of friction. A two-level modelling and control methodology is proposed, with the upper level using a friction-limited particle model for motion planning, and the lower level using a nonlinear 3DOF model for optimal control allocation. Motion planning adopts a two-phase approach: the first phase is to avoid the obstacle, the second is to recover lane keeping with minimal additional lateral deviation. This methodology differs from the more standard approach of path-planning/path-following, as there is no explicit path reference used; the control reference is a target acceleration vector which simultaneously induces changes in direction and speed. The lower level control distributes vehicle targets to the brake and steer actuators via a new and efficient method, the Modified Hamiltonian Algorithm (MHA). MHA balances CG acceleration targets with yaw moment tracking to preserve lateral stability. A nonlinear 7DOF two-track vehicle model confirms the overall validity of this novel methodology for collision avoidance.     

Emergency steering control of autonomous vehicle for collision avoidance and stabilisation

ABSTRACT

Collision avoidance and stabilisation are two of the most crucial concerns when an autonomous vehicle finds itself in emergency situations, which usually occur in a short time horizon and require large actuator inputs, together with highly nonlinear tyre cornering response. In order to avoid collision while stabilising autonomous vehicle under dynamic driving situations at handling limits, this paper proposes a novel emergency steering control strategy based on hierarchical control architecture consisting of decision-making layer and motion control layer. In decision-making layer, a dynamic threat assessment model continuously evaluates the risk associated with collision and destabilisation, and a path planner based on kinematics and dynamics of vehicle system determines a collision-free path when it suddenly encounters emergency scenarios. In motion control layer, a lateral motion controller considering nonlinearity of tyre cornering response and unknown external disturbance is designed using tyre lateral force estimation-based backstepping sliding-mode control to track a collision-free path, and to ensure the robustness and stability of the closed-loop system. Both simulation and experiment results show that the proposed control scheme can effectively perform an emergency collision avoidance manoeuvre while maintaining the stability of autonomous vehicle in different running conditions.     

Unified decision-making and control for highway collision avoidance using active front steer and individual wheel torque control

ABSTRACT

Collision avoidance is a crucial function for all ground vehicles, and using integrated chassis systems to support the driver presents a growing opportunity in active safety. With actuators such as in-wheel electric motors, active front steer and individual wheel brake control, there is an opportunity to develop integrated chassis systems that fully support the driver in safety critical situations. Here we consider the scenario of an impending frontal collision with a stationary or slower moving vehicle in the same driving lane. Traditionally, researchers have approached the required collision avoidance manoeuver as a hierarchical scheme, which separates the decision-making, path planning and path tracking. In this context, a key decision is whether to perform straight-line braking, or steer to change lanes, or indeed perform combined braking and steering. This paper approaches the collision avoidance directly from the perspective of constrained dynamic optimisation, using a single optimisation procedure to cover these aspects within a single online optimisation scheme of model predictive control (MPC). While the new approach is demonstrated in the context of a fully autonomous safety system, it is expected that the same approach can incorporate driver inputs as additional constraints, yielding a flexible and coherent driver assistance system.     

Risk management algorithm for rear-side collision avoidance using a combined steering torque overlay and differential braking

This paper describes a risk management algorithm for rear-side collision avoidance. The proposed risk management algorithm consists of a supervisor and a coordinator. The supervisor is designed to monitor collision risks between the subject vehicle and approaching vehicle in the adjacent lane. An appropriate criterion of intervention, which satisfies high acceptance to drivers through the consideration of a realistic traffic, has been determined based on the analysis of the kinematics of the vehicles in longitudinal and lateral directions. In order to assist the driver actively and increase driver's safety, a coordinator is designed to combine lateral control using a steering torque overlay by motor-driven power steering and differential braking by vehicle stability control. In order to prevent the collision while limiting actuator's control inputs and vehicle dynamics to safe values for the assurance of the driver's comfort, the Lyapunov theory and linear matrix inequalities based optimisation methods have been used. The proposed risk management algorithm has been evaluated via simulation using CarSim and MATLAB/Simulink.     

A nonlinear model predictive control formulation for obstacle avoidance in high-speed autonomous ground vehicles in unstructured environments

This paper presents a nonlinear model predictive control (MPC) formulation for obstacle avoidance in high-speed, large-size autono-mous ground vehicles (AGVs) with high centre of gravity (CoG) that operate in unstructured environments, such as military vehicles. The term ‘unstructured’ in this context denotes that there are no lanes or traffic rules to follow. Existing MPC formulations for passenger vehicles in structured environments do not readily apply to this context. Thus, a new nonlinear MPC formulation is developed to navigate an AGV from its initial position to a target position at high-speed safely. First, a new cost function formulation is used that aims to find the shortest path to the target position, since no reference trajectory exists in unstructured environments. Second, a region partitioning approach is used in conjunction with a multi-phase optimal control formulation to accommodate the complicated forms the obstacle-free region can assume due to the presence of multiple obstacles in the prediction horizon in an unstructured environment. Third, the no-wheel-lift-off condition, which is the major dynamical safety concern for high-speed, high-CoG AGVs, is ensured by limiting the steering angle within a range obtained offline using a 14 degrees-of-freedom vehicle dynamics model. Thus, a safe, high-speed navigation is enabled in an unstructured environment. Simulations of an AGV approaching multiple obstacles are provided to demonstrate the effectiveness of the algorithm.     

Optimal motion control for collision avoidance at Left Turn Across Path/Opposite Direction intersection scenarios using electric propulsion

Collision avoidance at intersections involving a host vehicle turning left across the path of an oncoming vehicle (Left Turn Across Path/Opposite Direction) have been studied in the past, but mostly using simplified interventions and rarely considering the possibility of crossing the intersection ahead of a bullet vehicle. Such a scenario where the driver preference is to avoid a collision by crossing the intersection ahead of a bullet vehicle is considered in this work. The optimal vehicle motion for collision avoidance in this scenario is determined analytically using a particle model within an optimal control framework. The optimal manoeuvres are then verified through numerical optimisations using a two-track vehicle model, where it was seen that the wheel forces followed the analytical global force angle result independently of the other wheels. A Modified Hamiltonian Algorithm controller for collision avoidance that uses the analytical optimal control solution is then implemented and tested in CarMaker simulations using a validated Volvo XC90 vehicle model. Simulation results showed that collision risk can be significantly reduced in this scenario using the proposed controller, and that more benefit can be expected in scenarios that require larger speed changes.     

Handling stability improvement through robust active front steering and active differential control

The design of the integrated active front steering and active differential control for handling improvement of road vehicles is undertaken. The controller design algorithm is based on the solution of a set of linear matrix inequalities that guarantee robustness against a number of vehicle parameters such as speed, cornering and braking stiffnesses. Vehicle plane dynamics are first expressed in the generic linear parameter-varying form, where the above-stated parameters are treated as interval uncertainties. Then, static-state feedback controllers ensuring robust performance against changing road conditions are designed. In a first series of simulations, the performance of the integrated controller is evaluated for a fishhook manoeuvre for different values of road adhesion coefficient. Then, the controller is tested for an emergency braking manoeuvre executed on a split-μ road. In all cases, it is shown that static-state feedback controllers designed by the proposed method can achieve remarkable road handling performance compared with uncontrolled vehicles.     

Pole location control design of an active suspension system with uncertain parameters

This paper addresses the problem of robust control design for an active suspension quarter-car model by means of state feedback gains. Specifically, the design of controllers that assure robust pole location of the closed-loop system inside a circular region on the left-hand side of complex plane is investigated. Three sufficient conditions for the existence of a robust stabilizing state feedback gain are presented as linear matrix inequalities: (i) the quadratic stability based gain; (ii) a recently published condition that uses an augmented space and has been here modified to cope with the pole location specification; (iii) a condition that uses an extended number of equations and yields a parameter-dependent state feedback gain. Unlike other parameter-dependent strategies, neither extensive gridding nor approximations are needed. In the suspension model, the sprung mass, the damper coefficient and the spring constant are considered as uncertain parameters belonging to a known interval (polytope type uncertainty). It is shown that the parameter-dependent gain proposed allows one to impose the closed-loop system pole locations that in some situations cannot be obtained with constant feedback gains.     

Pole location control design of an active suspension system with uncertain parameters

This paper addresses the problem of robust control design for an active suspension quarter-car model by means of state feedback gains. Specifically, the design of controllers that assure robust pole location of the closed-loop system inside a circular region on the left-hand side of complex plane is investigated. Three sufficient conditions for the existence of a robust stabilizing state feedback gain are presented as linear matrix inequalities: (i) the quadratic stability based gain; (ii) a recently published condition that uses an augmented space and has been here modified to cope with the pole location specification; (iii) a condition that uses an extended number of equations and yields a parameter-dependent state feedback gain. Unlike other parameter-dependent strategies, neither extensive gridding nor approximations are needed. In the suspension model, the sprung mass, the damper coefficient and the spring constant are considered as uncertain parameters belonging to a known interval (polytope type uncertainty). It is shown that the parameter-dependent gain proposed allows one to impose the closed-loop system pole locations that in some situations cannot be obtained with constant feedback gains.     

Model predictive control with constraints for a nonlinear adaptive cruise control vehicle model in transition manoeuvres

A vehicle following control law, based on the model predictive control method, to perform transition manoeuvres (TMs) for a nonlinear adaptive cruise control (ACC) vehicle is presented in this paper. The TM controller ultimately establishes a steady-state following distance behind a preceding vehicle to avoid collision, keeping account of acceleration limits, safe distance, and state constraints. The vehicle dynamics model is for continuous-time domain and captures the real dynamics of the sub-vehicle models for steady-state and transient operations. The ACC vehicle can execute the TM successfully and achieves a steady-state in the presence of complex dynamics within the constraint boundaries.     

A study on model fidelity for model predictive control-based obstacle avoidance in high-speed autonomous ground vehicles

This paper investigates the level of model fidelity needed in order for a model predictive control (MPC)-based obstacle avoidance algorithm to be able to safely and quickly avoid obstacles even when the vehicle is close to its dynamic limits. The context of this work is large autonomous ground vehicles that manoeuvre at high speed within unknown, unstructured, flat environments and have significant vehicle dynamics-related constraints. Five different representations of vehicle dynamics models are considered: four variations of the two degrees-of-freedom (DoF) representation as lower fidelity models and a fourteen DoF representation with combined-slip Magic Formula tyre model as a higher fidelity model. It is concluded that the two DoF representation that accounts for tyre nonlinearities and longitudinal load transfer is necessary for the MPC-based obstacle avoidance algorithm in order to operate the vehicle at its limits within an environment that includes large obstacles. For less challenging environments, however, the two DoF representation with linear tyre model and constant axle loads is sufficient.     

Designing H∞/GH 2 static-output feedback controller for vehicle suspensions using linear matrix inequalities and genetic algorithms

This paper presents an approach to design the H_∞/GH ₂ static-output feedback controller for vehicle suspensions by using linear matrix inequalities (LMIs) and genetic algorithms (GAs). Three main performance requirements for an advanced vehicle suspension are considered in this paper. Among these requirements, the ride-comfort performance is optimized by minimizing the H_∞ norm of the transfer function from the road disturbance to the sprung mass acceleration, while the road-holding performance and the suspension deflection limitation are guaranteed by constraining the generalized H₂ (GH ₂) norms of the transfer functions from the road disturbance to the dynamic tyre load and the suspension deflection to be less than their hard limits, respectively. At the same time, the controller saturation problem is considered by constraining its peak response output to be less than a given limit using the GH ₂ norm as well. A four-degree-of-freedom half-car model with active suspension system is applied in this paper. Several kinds of H_∞/GH ₂ static-output feedback controllers, which use the available sprung mass velocities or the suspension deflections as feedback signals, are obtained by using the GAs to search for the possible control gain matrices and then resolving the LMIs together with the minimization optimization problem. These designed H_∞/GH ₂ static-output feedback controllers are validated by numerical simulations on both the bump and the random road responses which show that the designed H_∞/GH ₂ static-output feedback controllers can achieve similar or even better active suspension performances compared with the state-feedback control case in spite of their simplicities.     

Design of robust-stable and quadratic finite-horizon optimal controllers with low trajectory sensitivity for uncertain active suspension systems

This paper presents a design method for designing the robust-stable and quadratic-finite-horizon-optimal controllers of uncertain active suspension systems. The method integrates a robust stabilisability condition, the orthogonal functions approach (OFA) and the hybrid Taguchi-genetic algorithm (HTGA). Using the integrative computational method, a robust-stable and quadratic-finite-horizon-optimal controller with low-trajectory sensitivity can be obtained such that (i) the active suspension system with elemental parametric uncertainties is stabilised and (ii) a quadratic-finite-horizon-integral performance index including a quadratic trajectory sensitivity term for the nominal active suspension system is minimised. The robust stabilisability condition is proposed in terms of linear matrix inequalities (LMIs). Based on the OFA, an algebraic algorithm only involving the algebraic computation is derived for solving the nominal active suspension feedback dynamic equations. By using the OFA and the LMI-based robust stabilisability condition, the dynamic optimisation problem for the robust-stable and quadratic-finite-horizon-optimal controller design of the linear uncertain active suspension system is transformed into a static-constrained-optimisation problem represented by the algebraic equations with constraint of LMI-based robust stabilisability condition; thus greatly simplifies the design problem. Then, for the static-constrained-optimisation problem, the HTGA is employed to find the robust-stable and quadratic-finite-horizon-optimal controllers of the linear uncertain active suspension systems. A design example is given to demonstrate the applicability of the proposed integrative computational approach.