用神经网络拟合车辆动力学是否要将车辆转移到车身坐标系内

This paper describes a neural network controller for autonomous underwater vehicles (AUVs). The designed online multilayer perceptron neural network (OMLPNN) calculates forces and moments in earth fixed frame to eliminate the tracking errors of AUVs whose dynamics are highly nonlinear and time varying. Another OMLPNN has been designed to generate an inverse model of AUV, which determine the appropriate propeller's speed and control surfaces' angles receiving the forces and moments in the body fixed frame. The designed approximation based neural network controller with the use of the backpropagation learning algorithm has advantages and robustness to control the highly nonlinear dynamics of AUV. The proposed neural networks architectures have been designed to control the test bed for AUV named NPS AUV. The Simulation results showed effectiveness of the OMLPNN to deal with elimination of AUVs' tracking errors as it has good capability to incorporate the dynamics of the system.

In this work, reinforcement learning is used to develop a position controller for an underactuated nature-inspired Unmanned Aerial Vehicle (UAV). This particular configuration of UAVs achieves lift by spinning its entire body contrary to standard multi-rotors or fixed-wing aircraft. Deep Deterministic Policy Gradients (DDPG) with Ape-X Distributed Prioritized Experience Replay was used to train neural network function approximators that were implemented as the final control policy. The reinforcement learning agent was trained in simulations and directly ported over to real-life hardware. Position control tests were performed on the learned control policy and compared to a baseline PID controller. The learned controller was found to exhibit better control over the inherent oscillations that arise from the non-linear dynamics of the platform.

The head-direction (HD) cells found in the limbic system in freely mov ing rats represent the instantaneous head direction of the animal in the horizontal plane regardless of the location of the animal. The internal direction represented by these cells uses both self-motion information for inertially based updating and familiar visual landmarks for calibration. Here, a model of the dynamics of the HD cell ensemble is presented. The stability of a localized static activity profile in the network and a dynamic shift mechanism are explained naturally by synaptic weight distribution components with even and odd symmetry, respectively. Under symmetric weights or symmetric reciprocal connections, a stable activity profile close to the known directional tuning curves will emerge. By adding a slight asymmetry to the weights, the activity profile will shift continuously without disturbances to its shape, and the shift speed can be controlled accurately by the strength of the odd-weight component. The generic formulation of the shift mechanism is determined uniquely within the current theoretical framework. The attractor dynamics of the system ensures modality-independence of the internal representation and facilitates the correction for cumulative error by the putative local-view detectors. The model offers a specific one-dimensional example of a computational mechanism in which a truly world-centered representation can be derived from observer-centered sensory inputs by integrating self-motion information.

11R9. Modeling and Simulation of Aerospace Vehicle Dynamics. AIAA Education Series. - PH Zipfel (Univ of Florida, Gainesville FL). AIAA, Reston VA. 2000. 551 pp. ISBN 1-56347-456-5. $79.95.Reviewed by W Schiehlen (Inst B of Mech, Univ of Stuttgart, Pfaffenwaldring 9, Stuttgart, 70550, Germany).This book can be used for one- and two-semester courses as well as for self-study. It presents the fundamentals of aerospace vehicle dynamics, and for simulations, the CADAC environment is available free of charge from the AIAA home page. Both components result in an excellent learning environment. The book starts with an overview of today’s computed-aided engineering concepts also called virtual engineering. Then, the two parts of the book Modeling of Flight Dynamics with six chapters and Simulation of Aerospace Vehicles with four chapters, are introduced. In the second chapter, the mathematical concepts of modeling are discussed based on classical mechanics and tensor calculus. Most important for aerospace vehicles are frames and coordinate systems which are presented in Chapter 3. The kinematics of translation and rotation are treated in Chapter 4 using the author’s concept of the rotational time derivative which has nice invariance properties. Then, in Chapter 5 the translational dynamics are considered based on Newtonian Dynamics and the corresponding transformations in moving reference frames. The rotational dynamics presented in Chapter 6 use Euler’s Law and includes also some aspects of gyrodynamics. The full 3D equations of motion are complex and not so easy to understand. In a first step, the perturbation techniques are introduced in Chapter 7 to get some insight in the linear behavior of aerospace vehicles. However, the results are restricted particular motions. Therefore, it is quite natural to proceed to the second part of the book. Chapter 8 is devoted to three-degree-of-freedom simulations with subsystem models for the atmosphere, the gravitational attraction, the drag, and the propulsion. Five-degree-of-freedom simulations follow in Chapter 9 with extended subsystems and more detailed missile simulations. Finally, in Chapter 10, six-degree-of-freedom simulations are introduced for flat Earth and elliptic Earth equations of motion. Subsystems for aerodynamics, autopilots, actuators, inertial navigation, guidance, and infrared sensors are added. The fundamentals of Monte Carlo simulations are presented and are well-written prototypes supporting simulations. The closing Chapter, 11, is related to real-time applications in flight simulators and hardware-in-the-loop facilities. The three appendices show some matrix calculus, the CADAC primer, and several trajectory simulations. Modeling and Simulation of Aerospace Vehicle Dynamics is very well written, and it is strongly recommended to students of aerospace engineering and to practitioners in industry. The author has applied his techniques widely in research and development to rockets, missiles, aircraft, and spacecraft. His experience is documented in the book. The examples and problems are helpful and clear.

Preface. 1. INTRODUCTION. 1.1 Systems and Control Terminology. 1.2 Nonlinear Systems. 1.3 Feedback Control Approaches. 1.3.1 Linear Design. 1.3.2 Adaptive Linear Design. 1.3.3 Nonlinear Design. 1.3.4 Adaptive Approximation Based Design. 1.3.5 Example Summary. 1.4 Components of Approximation Based Control. 1.4.1 Control Architecture. 1.4.2 Function Approximator. 1.4.3 Stable Training Algorithm. 1.5 Discussion and Philosophical Comments. 1.6 Exercises and Design Problems. 2. APPROXIMATION THEORY. 2.1 Motivating Example. 2.2 Interpolation. 2.3 Function Approximation. 2.3.1 Off-line (Batch) Function Approximation. 2.3.2 Adaptive Function Approximation. 2.4 Approximator Properties. 2.4.1 Parameter (Non)Linearity. 2.4.2 Classical Approximation Results. 2.4.3 Network Approximators. 2.4.4 Nodal Processors. 2.4.5 Universal Approximator. 2.4.6 Best Approximator Property. 2.4.7 Generalization. 2.4.8 Extent of Influence Function Support. 2.4.9 Approximator Transparency. 2.4.10 Haar Conditions. 2.4.11 Multivariable Approximation by Tensor Products. 2.5 Summary. 2.6 Exercises and Design Problems. 3. APPROXIMATION STRUCTURES. 3.1 Model Types. 3.1.1 Physically Based Models. 3.1.2 Structure (Model) Free Approximation. 3.1.3 Function Approximation Structures. 3.2 Polynomials. 3.2.1 Description. 3.2.2 Properties. 3.3 Splines. 3.3.1 Description. 3.3.2 Properties. 3.4 Radial Basis Functions. 3.4.1 Description. 3.4.2 Properties. 3.5 Cerebellar Model Articulation Controller. 3.5.1 Description. 3.5.2 Properties. 3.6 Multilayer Perceptron. 3.6.1 Description. 3.6.2 Properties. 3.7 Fuzzy Approximation. 3.7.1 Description. 3.7.2 Takagi-Sugeno Fuzzy Systems. 3.7.3 Properties. 3.8 Wavelets. 3.8.1 Multiresolution Analysis (MRA). 3.8.2 MRA Properties. 3.9 Further Reading. 3.10 Exercises and Design Problems. 4. PARAMETER ESTIMATION METHODS. 4.1 Formulation for Adaptive Approximation. 4.1.1 Illustrative Example. 4.1.2 Motivating Simulation Examples. 4.1.3 Problem Statement. 4.1.4 Discussion of Issues in Parametric Estimation. 4.2 Derivation of Parametric Models. 4.2.1 Problem Formulation for Full-State Measurement. 4.2.2 Filtering Techniques. 4.2.3 SPR Filtering. 4.2.4 Linearly Parameterized Approximators. 4.2.5 Parametric Models in State Space Form. 4.2.6 Parametric Models of Discrete-Time Systems. 4.2.7 Parametric Models of Input-Output Systems. 4.3 Design of On-Line Learning Schemes. 4.3.1 Error Filtering On-Line Learning (EFOL) Scheme. 4.3.2 Regressor Filtering On-Line Learning (RFOL) Scheme. 4.4 Continuous-Time Parameter Estimation. 4.4.1 Lyapunov Based Algorithms. 4.4.2 Optimization Methods. 4.4.3 Summary. 4.5 On-Line Learning: Analysis. 4.5.1 Analysis of LIP EFOL scheme with Lyapunov Synthesis Method. 4.5.2 Analysis of LIP RFOL scheme with the Gradient Algorithm. 4.5.3 Analysis of LIP RFOL scheme with RLS Algorithm. 4.5.4 Persistency of Excitation and Parameter Convergence. 4.6 Robust Learning Algorithms. 4.6.1 Projection modification. 4.6.2 &sigma -modification. 4.6.3 &epsis -modification. 4.6.4 Dead-zone modification. 4.6.5 Discussion and Comparison. 4.7 Concluding Summary. 4.8 Exercises and Design Problems. 5. NONLINEAR CONTROL ARCHITECTURES. 5.1 Small-Signal Linearization. 5.1.1 Linearizing Around an Equilibrium Point. 5.1.2 Linearizing Around a Trajectory. 5.1.3 Gain Scheduling. 5.2 Feedback Linearization. 5.2.1 Scalar Input-State Linearization. 5.2.2 Higher-Order Input-State Linearization. 5.2.3 Coordinate Transformations and Diffeomorphisms. 5.2.4 Input-Output Feedback Linearization. 5.3 Backstepping. 5.3.1 Second order system. 5.3.2 Higher Order Systems. 5.3.3 Command Filtering Formulation. 5.4 Robust Nonlinear Control Design Methods. 5.4.1 Bounding Control. 5.4.2 Sliding Mode Control. 5.4.3 Lyapunov Redesign Method. 5.4.4 Nonlinear Damping. 5.4.5 Adaptive Bounding Control. 5.5 Adaptive Nonlinear Control. 5.6 Concluding Summary. 5.7 Exercises and Design Problems. 6. ADAPTIVE APPROXIMATION: MOTIVATION AND ISSUES. 6.1 Perspective for Adaptive Approximation Based Control. 6.2 Stabilization of a Scalar System. 6.2.1 Feedback Linearization. 6.2.2 Small-Signal Linearization. 6.2.3 Unknown Nonlinearity with Known Bounds. 6.2.4 Adaptive Bounding Methods. 6.2.5 Approximating the Unknown Nonlinearity. 6.2.6 Combining Approximation with Bounding Methods. 6.2.7 Combining Approximation with Adaptive Bounding Methods. 6.2.8 Summary. 6.3 Adaptive Approximation Based Tracking. 6.3.1 Feedback Linearization. 6.3.2 Tracking via Small-Signal Linearization. 6.3.3 Unknown Nonlinearities with Known Bounds. 6.3.4 Adaptive Bounding Design. 6.3.5 Adaptive Approximation of the Unknown Nonlinearities. 6.3.6 Robust Adaptive Approximation. 6.3.7 Combining Adaptive Approximation with Adaptive Bounding. 6.3.8 Some Adaptive Approximation Issues. 6.4 Nonlinear Parameterized Adaptive Approximation. 6.5 Concluding Summary. 6.6 Exercises and Design Problems. 7. ADAPTIVE APPROXIMATION BASED CONTROL: GENERAL THEORY. 7.1 Problem Formulation. 7.1.1 Trajectory Tracking. 7.1.2 System. 7.1.3 Approximator. 7.1.4 Control Design. 7.2 Approximation Based Feedback Linearization. 7.2.1 Scalar System. 7.2.2 Input-State. 7.2.3 Input-Output. 7.2.4 Control Design Outside the Approximation Region D. 7.3 Approximation Based Backstepping. 7.3.1 Second Order Systems. 7.3.2 Higher Order Systems. 7.3.3 Command Filtering Approach. 7.3.4 Robustness Considerations. 7.4 Concluding Summary. 7.5 Exercises and Design Problems. 8. ADAPTIVE APPROXIMATION BASED CONTROL FOR FIXED-WING AIRCRAFT. 8.1 Aircraft Model Introduction. 8.1.1 Aircraft Dynamics. 8.1.2 Non-dimensional Coefficients. 8.2 Angular Rate Control for Piloted Vehicles. 8.2.1 Model Representation. 8.2.2 Baseline Controller. 8.2.3 Approximation Based Controller. 8.2.4 Simulation Results. 8.3 Full Control for Autonomous Aircraft. 8.3.1 Airspeed and Flight Path Angle Control. 8.3.2 Wind-axes Angle Control. 8.3.3 Body Axis Angular Rate Control. 8.3.4 Control Law and Stability Properties. 8.3.5 Approximator Definition. 8.3.6 Simulation Analysis. 8.4 Conclusions. 8.5 Aircraft Notation. Appendix A: Systems and Stability Concepts. A.1 Systems Concepts. A.2 Stability Concepts. A.2.1 Stability Definitions. A.2.2 Stability Analysis Tools. A.3 General Results. A.4 Prefiltering. A.5 Other Useful Results. A.5.1 Smooth Approximation of the Signum function. A.6 Problems. Appendix B: Recommended Implementation and Debugging Approach. References. Index.

No abstract

No abstract

Abstract This article develops a novel path‐following control strategy for underactuated unmanned surface vehicle (USV) subject to unmodeled dynamics and unknown multiple disturbance. A practical robust path‐following controller is proposed using trajectory linearization control (TLC) technology, neural network, and auxiliary design system. First, the greatest advantage of this article is that the TLC technology is first introduced into the field of USV motion control, which provides a new direction for TLC technology research. Second, the underactuated model based on a transformation of the USV kinematics to Serret‐Frenet frame is simplified by introducing a nonlinear coordinate transformation. Meanwhile, to improve the robustness and reduce the computational complexity, radial basis function neural network is replaced by neural network with minimum learning parameter method to compensate for unmodeled dynamics and unknown multiple disturbance. In addition, an auxiliary dynamic system is used to reduce the risk of actuator saturation. The stability of the whole system was proved based on the Lyapunov criteria. Finally, the comparison results demonstrate the superior performance of the proposed approach. © 2019 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

In this article, we investigate the prescribed-time fuzzy optimal tracking control for autonomous surface vehicles (ASVs). A monotonically decreasing boundary function that incorporates the settling time and tracking accuracy is proposed. A coordinate transformation on the boundary function and tracking error is proposed, and then an augmented system is defined. Subsequently, a new performance index function is presented that considers both the prescribed performance costs and control input costs. Given the inherent difficulties when directly resolving the Hamilton–Jacobi–Bellman equation within the prescribed-time framework, a new fuzzy optimal control scheme is proposed via integral reinforcement learning. This scheme does not require knowledge on the drift dynamics in the designed control policy and tuning laws, guarantees the simultaneous approximation of the optimal value function and control policy, ensures the stability of the ASV system, and allows users to specify the settling time and tracking accuracy. Finally, the presented strategy's effectiveness is validated by simulation.

This article proposes an adaptive critic attitude learning control for hypersonic morphing vehicles under large uncertainties and deformation. First, to remove the recursive design complexity caused by backstepping, an attitude-morphing coupled model in Brunovsky form is created by introducing a coordinate transformation. Second, a disturbance compensation controller is designed without using backstepping, wherein a resource-saving yet efficient unknown system dynamics estimator is introduced to estimate the lumped uncertainties by a simple filtering operation. Then, a near optimal regulator capable of online learning is developed under a critic-only adaptive dynamic programming framework. Notably, an improved finite-time updating law for critic weights is elaborated to achieve assured convergences by extracting weight errors from real-time and historical data. The significant merit is that even with fast morphing and strong uncertainties, good robustness, and optimal performances can be simultaneously attained under a convergence-assured learning setting. Through Lyapunov analysis, the convergences of the tracking error and weight estimation error are proven, guaranteeing the optimality of control strategies. Simulations are offered to demonstrate the advantages and utilities.

The full-fledged development and the practical use of autonomous vehicles (AVs) would be a great technological achievement and would substantially reduce the enormous damages caused by driving accidents to life and property. Technology based companies such as Google and Audi are getting closer to realizing the dream of seeing fully AVs on the road. A vehicle’s severely nonlinear dynamics due to the forces acting between road and vehicle tires, the coupling characteristic, and the uncertainties of parameters such as wheel moment of inertia and vehicle mass have made it rather difficult to approximate a precise mathematical model of vehicle dynamics. In this paper, to overcome these challenges we propose a model-independent control method based on improved adaptive neural controllers for path tracking control of AVs. In the structure of these improved neural controllers, we employ interval type-2 fuzzy sets (IT2FS) as activation functions. Despite the interdependence of a vehicle’s longitudinal and lateral motions, many of the research works on the path tracking of AVs have only focused on lateral motion control. By using the inputs of steering angle and torque, the presented control scheme tackles the simultaneous control of lateral and longitudinal moves. Results obtained from the lateral controller based on an improved neural network (NN) have been analyzed first at a constant velocity of 20 m/s and with/without considering parametric uncertainties. Then the longitudinal controller based on the improved NN is compared with sliding mode and common NN based controllers. Finally, the results obtained by simulating the simultaneous control of lateral and longitudinal motions indicate maximum tracking errors of 0.04 m (for lateral path following) and 0.02 m/s (for longitudinal velocity) and demonstrate the desirable performance of the proposed control approach.

This paper presents an appropriate method for estimating road friction coefficient. The method uses measured values from wheel angular velocity and yaw rate sensors of a vehicle so that it could estimate the road friction coefficient. The estimation process is done in three steps: first, vehicle lateral and longitudinal velocities along with yaw rate value are identified by an extended Kalman filter observer when lateral acceleration and yaw rate values are subjected to process and measurement noises, respectively. Then, lateral and longitudinal tire forces are estimated using a recursive least square algorithm so that to be used in a neural network designed based on well-known Magic Formula tire model. In the final stage, using a multilayer perceptron neural network and estimated values of the previous stages, the road friction coefficient is estimated. Finally, the set of estimators is evaluated using 14 degrees of freedom full vehicle dynamic model and the obtained results are compared with their actual values of vehicle model for two different maneuvers of vehicle.

Multibody models built in commercial software packages, e.g., ADAMS, can be used for accurate vehicle dynamics, but computational efficiency and numerical stability are very challenging in complex driving environments. These issues can be addressed by using data-driven models, owing to their robust generalization and computational speed. In this study, we develop a deep neural network (DNN) based model to predict longitudinal-lateral dynamics of an autonomous vehicle. Dynamic simulations of the autonomous vehicle are performed based on a semirecursive multibody method for data acquisition. The data are used to train and test the DNN model. The DNN inputs include the torque applied on wheels and the vehicle's initial speed that imitates a double lane change maneuver. The DNN outputs include the longitudinal driving distance, the lateral driving distance, the final longitudinal velocities, the final lateral velocities, and the yaw angle. The predicted vehicle states based on the DNN model are compared with the multibody model results. The accuracy of the DNN model is investigated in detail in terms of error functions. The DNN model is verified within the framework of a commercial software package CarSim. The results demonstrate that the DNN model predicts accurate vehicle states in real time. It can be used for real-time simulation and preview control in autonomous vehicles for enhanced transportation safety.

Accurate estimation of vehicle sideslip angle and attitude angles are essential for the safety control and lateral behaviour of driving performance. In this article, the variation of wheels cornering stiffness is considered for sideslip estimation and addressed by introducing a recursive least squares approach. Based on the nonlinear vehicle dynamic model and the investigated coupling effect between lateral and longitudinal velocity, an optimized moving horizon estimator is proposed to obtain the vehicle sideslip angle, in which an iteration decent algorithm is integrated. Furthermore, a framework, consisting of inertial navigation system measurements, a dual neural network and a square-root cubature Kalman filter, is designed, such that the influence of sensor noise and varied maneuvers are alleviated when estimating the system states. Finally, extensive simulation and field experiments are carried out on different driving scenarios to verify the effectiveness of the developed method. The obtained results clearly indicate the satisfactory estimation accuracy of the designed strategy, superior to the existing estimation methods, such as sole neural networks methods and Kalman-based filters.

Vehicle dynamics control systems have a fundamental role in smart and autonomous mobility, where one of the most crucial aspects is the vehicle body velocity estimation. In this paper, the problem of a correct evaluation of the vehicle longitudinal velocity for dynamic control applications is approached using a neural networks technique employing a set of measured samples referring to signals usually available on-board, such as longitudinal and lateral acceleration, steering angle, yaw rate and linear wheel speed. Experiments were run on four professional driving circuits with very different characteristics, and the vehicle longitudinal velocity was estimated with different neural network training policies and validated through comparison with the measurements of the one acquired at the vehicle's center of gravity, provided by an optical Correvit sensor, which serves as the reference (and, therefore, exact) velocity values. The results obtained with the proposed methodology are in good agreement with the reference values in almost all tested conditions, covering both the linear and the nonlinear behavior of the car, proving that artificial neural networks can be efficiently employed onboard, thereby enriching the standard set of control and safety-related electronics.

Accurate estimation of the vehicle sideslip angle is fundamental in vehicle dynamics control and stability. In this paper two different methods for vehicle sideslip estimation, based on Principal Component Analysis (PCA) and Neural Networks (NN), are presented comparing the procedure responses with full-scale vehicle acquired test data. The estimation algorithms use driver's steering angle, lateral and longitudinal accelerations, wheel angular velocities and yaw rate measured from sensors integrated in a test vehicle, and are validated by comparison with the measurements of the sideslip angle provided by an optical Correvit sensor suitably mounted on board, serving as the reference system in terms of accuracy of slip-free measurement of longitudinal and transverse vehicle dynamics. The procedure results, based on both the original (RAW) and the reduced (PCA) data sets, are compared to the acquired sideslip angle, using the estimated channel as an input for the TRICK tool to evaluate the accuracy of the results and the potential of the estimation process in terms of tire interaction curves.

Active safety systems, such as the electronic stability control (ESC), have been widely utilized in modern vehicles. The feedback control of these systems requires accurate vehicle states information such as vehicle longitudinal velocity and sideslip angle. In this article, a novel vehicle dynamics estimator based on adaptive Kalman filter utilizing long short-term memory neural network (LSTM-AKF) is proposed to observe longitudinal velocity and sideslip angle. A planar vehicle kinematics model is adopted for constructing the state-space equation of the LSTM-AKF. Two virtual sensors are designed to obtain the longitudinal and lateral measurements for the LSTM-AKF, respectively. The first virtual sensor is designed for longitudinal measurements, utilizing wheel speeds and acceleration signals to estimate the longitudinal velocity. The second virtual sensor, based on an LSTM neural network, estimates lateral velocity to derive the sideslip angle. To enhance the robustness of state observation, the LSTM-AKF estimator considers the uncertainties of measurements by incorporating measurement noise covariance adjustments through two LSTM neural networks. The training labels for these two networks are designed based on a feedback method. A dataset based on real vehicle experiments is constructed to train the networks. Finally, the performance of the LSTM-AKF estimator is examined through longitudinal and lateral field test scenarios under high-adhesion and low-adhesion roads. The results demonstrate that the LSTM-AKF estimator exhibits higher estimation accuracy compared with baseline methods.

<div>One of the most important factors affecting the performance of vehicle active chassis control systems is the tire-road friction coefficient. Accurate estimation of the friction coefficient can lead to better performance of these controllers. In this study, a new three-step friction estimation algorithm, based on intelligent tire concept, is proposed, which is a combination of experiment-based and vehicle dynamic based approaches. In the first step of the proposed algorithm, the normal load is estimated using a trained Artificial Neural Network (ANN). The network was trained using the experimental data collected using a portable tire testing trailer. In the second step of the algorithm, the tire forces and the wheel longitudinal velocity are estimated through a two-step Kalman filter. Then, in the last step, using the estimated tire normal load and longitudinal and lateral forces, the friction coefficient can be estimated. To evaluate the performance of the algorithm, experiments were performed using the trailer test setup and friction was calculated using the measured forces. Good agreement was observed between the estimated and actual friction coefficients.</div>

In the last few decades, the role of vehicle dynamics control systems has become crucial. In this complex scenario, the correct real-time estimation of the vehicle’s sideslip angle is decisive. Indeed, this quantity is deeply linked to several aspects, such as traction and stability optimization, and its correct understanding leads to the possibility of reaching greater road safety, increased efficiency, and a better driving experience for both autonomous and human-controlled vehicles. This paper aims to estimate accurately the sideslip angle of the vehicle using different neural network configurations. Then, the proposed approach involves using two separate neural networks in a dual-network architecture. The first network is dedicated to estimating the longitudinal velocity, while the second network predicts the sideslip angle and takes the longitudinal velocity estimate from the first network as input. This enables the creation of a virtual sensor to replace the real one. To obtain a reliable training dataset, several test sessions were conducted on different tracks with various layouts and characteristics, using the same reference instrumented vehicle. Starting from the acquired channels, such as lateral and longitudinal acceleration, steering angle, yaw rate, and angular wheel speeds, it has been possible to estimate the sideslip angle through different neural network architectures and training strategies. The goodness of the approach was assessed by comparing the estimations with the measurements obtained from an optical sensor able to provide accurate values of the target variable. The obtained results show a robust alignment with the reference values in a great number of tested conditions. This confirms that the adoption of artificial neural networks represents a reliable strategy to develop real-time virtual sensors for onboard solutions, expanding the information available for controls.

The majority of currently used automatic parking systems exploit the planning-and-tracking approach that involves planning the reference trajectory first and then tracking the desired reference trajectory. However, the response delay of longitudinal velocity prevents the parking controller from tracing the desired trajectory because the vehicle’s velocity and other state parameters are not synchronized, while the controller maneuvers the vehicle according to the planned desired velocity and steering profiles. We propose an inverse vehicle model to provide a neural-network-based integrated lateral and longitudinal automatic parking controller. We approximated the relationship of the planned velocity to the vehicle’s velocity using a second-order difference equation that involves the response characteristic of the vehicle’s longitudinal delay. The adjusted desired velocity to track the origin-planned velocity is calculated using the inverse vehicle model. Furthermore, we proposed an integrated longitudinal and lateral parking controller using an artificial neural network (ANN) model trained on a dataset applying the inverse vehicle model. By learning the control laws between the vehicle’s states and the corresponding actions, the proposed ANN-based controller could yield a steering angle and the adjusted desired velocity to complete automatic parking in a confined space.

<div class="section abstract"><div class="htmlview paragraph">The growing market demand for highly automated and autonomous vehicles and the need to equip vehicles with ever higher standards of comfort, safety and performance requires knowledge of physical quantities that are often difficult or expensive to measure directly. The absence of direct sensors, the difficulty of implementation, and their cost have led researchers to identify alternative solutions that allow estimating the physical quantity of interest by aggregating other available information. The interaction forces between tire and road are among the most significant. Given that the dynamics of a vehicle are strongly linked to the forces exchanged between the tire and the road, their knowledge is fundamental in the development of control systems aimed at improving performance in terms of handling, road holding or comfort. This paper presents a new technique for the estimation of tire-road interaction forces based on the integration of models and measures. A Central Difference Kalman filter was applied to a Double Track Model. The non-linear Kalman filter allowed us to handle the non-linearity of the system. The tire-road interaction was modelled through Pacejka's magic formulas that into account the combined longitudinal and lateral slips and the camber angle. This version made it possible to carry out complex and realistic manoeuvres. The realized estimator also considers the influence of lateral and longitudinal load transfers and aerodynamic forces in the three spatial directions. The Camber angle used in this observer was estimated through neural networks. The measures used are longitudinal velocity, yaw rate, longitudinal slip and wheel steering angles.</div></div>

In order to drive safely and efficiently on public roads, autonomous vehicles will have to understand the intentions of surrounding vehicles, and adapt their own behavior accordingly. If experienced human drivers are generally good at inferring other vehicles' motion up to a few seconds in the future, most current Advanced Driving Assistance Systems (ADAS) are unable to perform such medium-term forecasts, and are usually limited to high-likelihood situations such as emergency braking. In this article, we present a first step towards consistent trajectory prediction by introducing a long short-term memory (LSTM) neural network, which is capable of accurately predicting future longitudinal and lateral trajectories for vehicles on highway. Unlike previous work focusing on a low number of trajectories collected from a few drivers, our network was trained and validated on the NGSIM US-101 dataset, which contains a total of 800 hours of recorded trajectories in various traffic densities, representing more than 6000 individual drivers.

<div class="section abstract"><div class="htmlview paragraph">As an important input parameter of intelligent vehicle active safety technology, road adhesion coefficient is of great significance in autonomous collision avoidance, emergency braking and collision avoidance, and variable adhesion road motion control. Traditional recognition methods based on vehicle dynamics require large data volume and low solution accuracy. This paper proposes an adhesion coefficient recognition method based on Elman neural network and Kalman filter. By establishing a seven-degree-of-freedom vehicle dynamics model, dynamic parameters such as yaw angular velocity, longitudinal velocity, lateral velocity, and angular velocity of each wheel, which are easy to measure and strongly related to the road adhesion coefficient, are analyzed as the input of the neural network model. The square root cubature Kalman filter algorithm is used to remove the noise of the input of the neural network model, and Q-learning is used to strengthen learning, and the weight coefficient and bias coefficient of the model are continuously rewarded and punished, so that the predicted value does not exceed the normal range of values. The algorithm was pre-trained through CarSim/Simulink co-simulation, 9 sets of simulation conditions were established, and 4 sets of verification schemes were designed for identification and inspection. The average error of the simulation process is 4.93%, and the accuracy is 91.23%. Compared with the traditional Elman neural network, the average recognition error of this method is reduced by 2.23%, and the accuracy rate is increased by 9.83%. Real vehicle experiments were carried out on wet asphalt pavement and dry asphalt pavement, verifying the feasibility of the method. This paper proposes a road adhesion coefficient recognition method, which can improve the applicability of intelligent vehicle active safety systems to complex scenarios.</div></div>

Preceding target vehicle (PTV) motion recognition play a pivotal role in autonomous vehicles. Motion states such as yaw rate, longitudinal and lateral velocity are critical for ego vehicle decision-making and control. However, lateral states of a PTV can hardly be measured directly by common onboard sensors and the PTV lateral state estimation has been seldom addressed in existing literatures. In this paper, a novel estimation scheme based on multiple neural network ensemble is proposed for PTV lateral state estimation. First, PTV lateral kinematics is presented based on vehicle-road relationship and a novel PTV lateral motion model is constructed to interpret the PTV lateral motion. Then, neural network observer with the PTV lateral kinematics as prior knowledge is designed and training data are collected in simulation environment. The neural network observer is trained using Levenberg-Marquardt backpropagation with Bayesian regularization (LMBR) to improve the generalization capability. Finally, to further improve the performance of the neural network estimation method, multiple neural network observers are integrated by weighted averaging strategy. The effectiveness of proposed approach is verified through hardware-in-the-Ioop (HiL) experiments conducted in designed verification scenarios, and compared with model-based method and other three learning methods. The experiment results reveal that the proposed method outperforms other typical methods and achieves accurate estimation of the PTV lateral states.

Model Predictive Controller (MPC) is widely used as a technique for path tracking control since it allows for dealing with system constraints and future forecasts. However, the performance of MPC is directly affected by the adopted model. A complex dynamic model can guarantee accuracy in path tracking but may not be suitable in computational terms. On the other hand, a simplified model may not capture essential nonlinear aspects. Thus, to cope with these problems, this paper deals with data-driven tire modeling to improve autonomous ground vehicle path tracking control. The main contribution of the present work is to show that neural tires can capture the nonlinearities present in the interaction between lateral and longitudinal vehicle dynamics, with a reduced computational cost for predictive controllers. Simulated and experimental tire data are approximate to design data-driven tire models using radial basis function and multilayer perceptron neural networks. Then, based on ground vehicles with neural tires, model predictive controllers are designed to regulate wheel torque and steering angle inputs. Comparative tests were conducted to compare the proposed data-driven MPC approach with the classical nonlinear MPC controller. The results show that the neural tires approximate nonlinear tire models and experimental data with arbitrary precision in terms of accuracy and error-based metrics. The proposed methodology was successfully applied to perform trajectory and velocity tracking of ground vehicles. In addition, MPC with a neural tire model as prediction inference reduces the computational effort compared to traditional approaches.

Effective estimation of vehicle states such as the yaw rate and the sideslip angle is important for vehicle stability control. Unfortunately the devices are very expensive to measure the sideslip angle directly and are not suitable for ordinary vehicle. Therefore, it must be estimated. A novel sideslip angle soft-sensor using neural network left inversion (NNLI) is presented for the in-wheel motor driven electric vehicle (EV). The innovation of the presented algorithm is not only little concerned with reference model parameters identification, but also uses the characteristic of the in-wheel motor driven EV. Longitudinal acceleration, lateral acceleration, yaw rate, longitudinal velocity, steering angle, the torque of in-wheel motor which can be acquired by ordinary sensors are used as inputs. Co-simulations are carried out to demonstrate the effectiveness of the proposed soft-sensor with Simulink and CarSim.

<div class="section abstract"><div class="htmlview paragraph">The road adhesion coefficient has a great impact on the performance of vehicle tires, which in turn affects vehicle safety and stability. A low coefficient of adhesion can significantly reduce the tire's traction limit. Therefore, the measurement of the coefficient is much helpful for automated vehicle control and stability control. Considering that the road adhesion coefficient is an inherent parameter of the road and it cannot be known directly from the information of the on-vehicle sensors. The novelty of this paper is to construct a road adhesion coefficient observer which considers the noise of sensors and measures the unknown state variable by the trained neural network. A Butterworth filter and Adaptive Neural Fuzzy Interference System (ANFIS) are combined to provide the lateral and longitudinal velocity which cannot be measured by regular sensors. Unscented Kalman filter (UKF) considering vehicle model, wheel model, and tire model is proposed to estimate the road adhesion coefficient. Eventually, the road adhesion coefficient observer is tested in some varying road conditions and compared with some typical methods to verify the feasibility of the estimation theory proposed in this paper.</div></div>

The increasing demand for reliable and efficient operation of Permanent Magnet Synchronous Motors (PMSMs) in electric vehicle (EV) motor systems necessitates advanced real-time fault detection techniques. This study presents a Hybrid Digital Twin (HDT) framework that integrates Physics-Informed Neural Networks (PINNs) and an Extended Kalman Filter (EKF)-based sensor fusion strategy to enable accurate, adaptive, and real-time condition monitoring and fault diagnosis of PMSMs. Initially, motor signals are transformed into the rotating reference frame using the dq-axis transformation to facilitate simplified modeling. The PINN model, informed by fundamental motor physics, estimates stator flux linkages from voltage and current inputs. These estimates are refined by the EKF, which mitigates sensor noise and model uncertainties, enhancing the fidelity of state estimation. Residuals between measured and predicted signals are analyzed, and a Support Vector Machine (SVM) classifier is used to detect and categorize faults. ►The proposed HDT framework achieves a fault detection accuracy of 96.99%, outperforming traditional Motor Current Signature Analysis (MCSA) by approximately 13%. The model also achieves high precision (90.09%), recall (89.21%), and F1-score (89.65%), with a fault detection latency of under 95 milliseconds and a 31% reduction in false positives. The combined use of physics-based flux estimation, real-time EKF correction, residual analysis, and FFT-based spectral features ensures reliable and interpretable diagnostics. Overall, the HDT framework offers a scalable and robust solution for real-time PMSM fault detection in EV motor applications, particularly under dynamic load conditions. • Hybrid digital twin detects PMSM faults in EVs with 96.99% accuracy. • Real-time fault diagnosis achieved in 95 ms with 33% fewer false alarms. • PINNs reduce need for large labeled data, enabling scalable models. • RL-based threshold tuning boosts accuracy under dynamic conditions. • dq-transform, FFT, and EKF fusion ensure robust time-frequency analysis.

A multitude of cyber-physical system (CPS) applications, including design, control, diagnosis, prognostics, and a host of other problems, are predicated on the assumption of model availability. There are mainly two approaches to modeling: Physics/Equation based modeling (Model-Based, MB) and Machine Learning (ML). Recently, there is a growing consensus that ML methodologies relying on data need to be coupled with prior scientific knowledge (or physics, MB) for modeling CPS. We refer to the paradigm that combines MB approaches with ML as hybrid learning methods. Hybrid modeling (HB) methods is a growing field within both the ML and scientific communities, and are recognized as an important emerging but nascent area of research. Recently, several works have attempted to merge MB and ML models for the complete exploitation of their combined potential. However, the research literature is scattered and unorganized. So, we make a meticulous and systematic attempt at organizing and standardizing the methods of combining ML and MB models. In addition to that, we outline five metrics for the comprehensive evaluation of hybrid models. Finally, we conclude by shedding some light on the challenges of hybrid models, which we, as a research community, should focus on for harnessing the full potential of hybrid models. An additional feature of this survey is that the hybrid modeling work has been discussed with a focus on modeling cyber-physical systems.

No abstract

Model Predictive Controller (MPC) is a capable technique for designing Path Tracking Controller (PTC) of Autonomous Vehicles (AVs). The performance of MPC can be significantly enhanced by adopting a high-fidelity and accurate vehicle model. This model should be capable of capturing the full dynamics of the vehicle, including nonlinearities and uncertainties, without imposing a high computational cost for MPC. A data-driven approach realised by learning vehicle dynamics using vehicle operation data can offer a promising solution by providing a suitable trade-off between accurate state predictions and the computational cost for MPC. This work proposes a framework for designing an MPC with a Neural Network (NN)-based learned dynamic model of the vehicle using the plethora of data available from modern vehicle systems. The objective is to integrate an NN-based model with higher accuracy than the conventional vehicle models for the required prediction horizon into MPC for improved tracking performances. The proposed NN-based model is highly capable of approximating latent system states, which are difficult to estimate, and provides more accurate predictions in the presence of parametric uncertainties. The observation of the results in various road conditions shows that the proposed approach outperforms the MPCs with conventional vehicle models.

This article describes some technical issues regarding the adaptation of a production car to a platform for the development and testing of autonomous driving technologies. A universal approach to performing the reverse engineering of electric power steering (EPS) for the purpose of external control is also presented. The primary objective of the related study was to solve the problem associated with the precise prediction of the dynamic trajectory of an autonomous vehicle. This was accomplished by deriving a new equation for determining the lateral tire forces and adjusting some of the vehicle parameters under road test conductions. A Mivar expert system was also integrated into the control system of the experimental autonomous vehicle. The expert system was made more flexible and effective for the present application by the introduction of hybrid artificial intelligence with logical reasoning. The innovation offers a solution to the major problem of liability in the event of an autonomous transport vehicle being involved in a collision.

Vehicle sideslip angle is a major indicator of dynamics stability for ground vehicles; but it is immeasurable with commercially‐available sensors. Sideslip angle estimation has been the focus of intensive research in past decades, resulting in a rich library of related literature. This study presents a comprehensive evaluation of state‐of‐the‐art sideslip angle estimation methods, with the primary goal of quantitatively revealing their strengths and limitations. These include kinematics‐, dynamics‐ and neural network‐based estimators. A hardware‐in‐loop system is purposely established to examine their performance under four typical manoeuvres. The results show that the dynamics‐based estimators are suitable at low vehicle velocities when tires operate in the linear region. In contrast, the kinematics‐based methods yield superior estimation performance at high vehicle velocities, and the inclusion of the dual GPS receivers is beneficial even when there is large disturbance to the steering angle. Of utmost importance, it is experimentally manifested that the neural network‐based estimator can perform well in all manoeuvres once the training datasets are properly selected.

No abstract

In this paper, an optimal control strategy for a four-wheel-independently-actuated electric vehicle (FWIA EV) is proposed to improve vehicle dynamics stability and handling performance. The proposed scheme has a hierarchical structure composed of an upper and a lower controller. The desired longitudinal and lateral forces and yaw moment are determined based on the sliding-mode control (SMC) scheme in the upper controller, which takes the longitudinal and lateral velocity and the yaw rate as control variables. In the lower controller, an optimization algorithm is adopted to allocate the driving/braking torques to each in-wheel motor. A cost function with adjustable weight coefficients is specially designed by taking the motor power capability and the tire workload into consideration. The simulation and hardware-in-loop experimental results show that the proposed control strategy exhibits superior performance in comparison to commonly-used rule-based control strategies, and has the capability of online implementation.

No abstract