No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Model Free Adaptive Control Algorithm based on ReOSELM for Autonomous Driving Vehicles
... In terms of the experimental hazards of emergency circumstances, the related work was validated in the MeiTuan autonomous vehicle onboard simulation system with the virtual environment and embedded vehicle model [6], as shown in Fig. 1. Corresponding simulation data are used to evaluate this time-optimal trajectory planning method according to the minimum passing time and dynamic responses of vehicles in emergency scenarios. ...
... Owing to this, the small steering angle assumption is no longer applicable here, and the original maximum longitudinal speed restriction (3) in terms of lateral friction limits is not eligible. At this point, the accurate formulation considering steering angle is (6). Then the longitudinal speed should be restricted by (7). ...
... Fig. 2(a), the road friction coefficient in this scenario is 0.3, and the vehicle parameters are given by Table. 1. In this section, the trajectory planning with original simplified longitudinal speed limits and corrected longitudinal speed limits are simulated and compared in the Meituan autonomous vehicle onboard simulation system [6], respectively. ...
In emergency circumstances, it is essential for autonomous vehicles to balance stability and dynamic performance to attain a faster travel speed while preserving stability. It is not unusual to find traffic accidents caused by suddenly present intruders on the road. In this situation, if there is not enough distance for the vehicle to brake immediately, the vehicle needs to operate with a relatively big steering angle and cornering speed to avoid collision while maintaining driving stability. This can be a challenging scenario even for a human driver, let alone autonomous driving. Especially, this poses a burden on trajectory optimization. In this case, neither over-conservative nor unachievable trajectory and speed profiles are eligible. Technically, the difficulty lies in an accurate maximum cornering speed estimation due to the impact of nonlinear tire force responses in these scenarios with large steering angles and high cornering speed. While this difficulty can be addressed by introducing tire model and extra variables, like tire stiffness and shape factor, in the formulation of this problem to cover nonlinear effect, it ends up increasing the complexity of the model and optimization problem. In this paper, a novel maximum curve-passing speed correction method for online trajectory optimization is proposed by leveraging predefined nonlinear correction terms, which is applicable without introducing any extra variables to the optimization. Moreover, this method has been simulated in the autonomous vehicle software-in-the-loop system. It is validated that this method can achieve online trajectory planning with maximum curve-passing speed while ensuring lateral stability.
For large-scale networked plant-wide systems composed by physically (or geographically) divided subsystems, only limited information is available for local controllers on account of region and communication restrictions. Concerning the optimal control problem of such subsystems, a neighbor-based distributed model predictive control (NDMPC) strategy is presented to improve the global system performance. In this scheme, the performance index of local subsystems and that of its neighbors are minimized together in the determination of the optimal control input, which makes the local control decision also beneficial to its neighboring subsystems and further contributes to improving the convergence and control performance of overall system. The stability of the closed-loop system is proved. Moreover, the parameter designing method for distributed synthesis is provided. Finally, the simulation results illustrate the main characteristics and effectiveness of the proposed control scheme.
This work proposes a framework to design, formulate and implement a path tracker for self‐driving cars (SDC) based on a nonlinear model‐predictive‐control (NMPC) approach. The presented methodology is developed to be used by designers in the industrial sector, practitioners, and academics. Therefore, it is straight forward, flexible, and comprehensive. It allows the designer to easily integrate multiple objective terms in the cost function either opposing or correlating. The proposed design of the controller not only targets accurate tracking but also comfortable ride and fast travel time by introducing several sub‐objective terms in the main cost function to satisfy these goals. These sub‐objective terms are weighted according to their contribution to the optimization problem. The SDC‐NMPC framework is developed using the high‐performance language C++ and utilizes highly optimized math and optimization libraries for best real‐time performance. This makes the SDC‐MPC well suited for use in both ADAS and self‐driving cars. Extensive simulation studies featuring complex tracks with many sharp turns have been carried out to evaluate the performance of the proposed SDC‐NMPC at different speeds. The presented analysis shows that the proposed controller with its tuning technique outperforms that of the PID‐based controller.
Recently, model predictive control (MPC) is increasingly applied to path tracking of mobile devices, such as mobile robots. The characteristics of these MPC-based controllers are not identical due to the different approaches taken during design. According to the differences in the prediction models, we believe that the existing MPC-based path tracking controllers can be divided into four categories. We named them linear model predictive control (LMPC), linear error model predictive control (LEMPC), nonlinear model predictive control (NMPC), and nonlinear error model predictive control (NEMPC). Subsequently, we built these four controllers for the same mobile robot and compared them. By comparison, we got some conclusions. The real-time performance of LMPC and LEMPC is good, but they are less robust to reference paths and positioning errors. NMPC performs well when the reference velocity is high and the radius of the reference path is small. It is also robust to positioning errors. However, the real-time performance of NMPC is slightly worse. NEMPC has many disadvantages. Like LMPC and LEMPC, it performs poorly when the reference velocity is high and the radius of the reference path is small. Its real-time performance is also not good enough.
In this paper, we propose a new design method of discrete-valued model predictive control for continuous-time linear time-invariant systems based on sum-of-absolute-values (SOAV) optimization. The finite-horizon discrete-valued control design is formulated as an SOAV optimal control, which is an expansion of L¹ optimal control. It is known that under the normality assumption, the SOAV optimal control exists and takes values in a fixed finite alphabet set if the initial state lies in a subset of the reachable set. In this paper, we analyze the existence and discreteness property for systems that do not necessarily satisfy the normality assumption. Then, we extend the finite-horizon SOAV optimal control to infinite-horizon model predictive control (MPC). We give sufficient conditions for the recursive feasibility and the stability of the MPC-based feedback system in the presence of bounded noise. Simulation results show the effectiveness of the proposed method.
A simple first-order discrete-time nonlinear system, which has both parametric uncertainty and non- parametric uncertainty, is studied in this paper. The uncertainty of non-parametric part is characterized by a Lipschitz constant L, and the nonlinearity of parametric part is characterized by an exponent index b. An adaptive controller is constructed for this model in both cases of b = 1 and b > 1, and its closed-loop stability is established under some conditions. When b = 1, the conditions given reveal the magic number 3/2 + radic2 which appeared in previous study on capability and limitations of the feedback mechanism.
In this work, a novel data-driven control approach, model-free adaptive control, is presented based on a new dynamic linearization technique for a class of discrete-time single-input and single-output nonlinear systems. The main feature of the approach is that the controller design depends merely on the input and the output measurement data of the controlled plant. The theoretical analysis shows that the approach guarantees the bounded input and bounded output stability and tracking error monotonic con- vergence. The comparison experiments verify the effectiveness of the proposed approach. Index Terms—Adaptive control, compact form dynamic lin- earization (CFDL), data-driven control, model-free adaptive control (MFAC), partial form dynamic linearization (PFDL), pseudo-partial derivative (PDD), three-tank system.
According to conventional neural network theories, single-hidden-layer feedforward networks (SLFNs) with additive or radial basis function (RBF) hidden nodes are universal approximators when all the parameters of the networks are allowed adjustable. However, as observed in most neural network implementations, tuning all the parameters of the networks may cause learning complicated and inefficient, and it may be difficult to train networks with nondifferential activation functions such as threshold networks. Unlike conventional neural network theories, this paper proves in an incremental constructive method that in order to let SLFNs work as universal approximators, one may simply randomly choose hidden nodes and then only need to adjust the output weights linking the hidden layer and the output layer. In such SLFNs implementations, the activation functions for additive nodes can be any bounded nonconstant piecewise continuous functions g : R --> R and the activation functions for RBF nodes can be any integrable piecewise continuous functions g : R --> R and integral of R g(x)dx not equal to 0. The proposed incremental method is efficient not only for SFLNs with continuous (including nondifferentiable) activation functions but also for SLFNs with piecewise continuous (such as threshold) activation functions. Compared to other popular methods such a new network is fully automatic and users need not intervene the learning process by manually tuning control parameters.
It is clear that the learning speed of feedforward neural networks is in general far slower than required and it has been a major bottleneck in their applications for past decades. Two key reasons behind may be: 1) the slow gradient-based learning algorithms are extensively used to train neural networks, and 2) all the parameters of the networks are tuned iteratively by using such learning algorithms. Unlike these traditional implementations, this paper proposes a new learning algorithm called extreme learning machine (ELM) for single-hidden layer feedforward neural networks (SLFNs) which randomly chooses the input weights and analytically determines the output weights of SLFNs. In theory, this algorithm tends to provide the best generalization performance at extremely fast learning speed. The experimental results based on real-world benchmarking function approximation and classification problems including large complex applications show that the new algorithm can produce best generalization performance in some cases and can learn much faster than traditional popular learning algorithms for feedforward neural networks.
In this paper, an intelligent automated lane-keeping system is proposed and implemented on our vehicle platform, i.e., TAIWAN i TS-1. This system challenges the online integrating heterogeneous systems such as a real-time vision system, a lateral controller, in-vehicle sensors, and a steering wheel actuating motor. The implemented vision system detects the lane markings ahead of the vehicle, regardless of the varieties in road appearance, and determines the desired trajectory based on the relative positions of the vehicle with respect to the center of the road. To achieve more humanlike driving behavior such as smooth turning, particularly at high levels of speed, a fuzzy gain scheduling (FGS) strategy is introduced to compensate for the feedback controller for appropriately adapting to the SW command. Instead of manual tuning by trial and error, the methodology of FGS is designed to ensure that the closed-loop system can satisfy the crossover model principle. The proposed integrated system is examined on the standard testing road at the Automotive Research and Testing Center (ARTC)<sup>1</sup> and extra-urban highways.
One of the most critical obstacles in the automated operation of
commercial heavy vehicles (CHVs) is the presence of significant delays
in the fuel and brake actuators. These delays are especially important
in longitudinal control of vehicle platoons that do not employ
intervehicle communication, because their effect becomes cumulative as
it propagates upstream, resulting in considerably degraded performance.
Our objective in this paper is to design autonomous controllers which,
in the presence of large delays, recover the good performance achieved
when the delays are negligible. We propose two different approaches
which are tailored to different performance requirements and
computational resources. A backstepping-based nonlinear scheme with
prediction almost recovers the original “delay-free”
performance at the cost of additional controller complexity, while a
simpler PID-based nonlinear scheme yields nearly as good performance
This paper investigates the vision-based autonomous driving with deep learning and reinforcement learning methods. Different from the end-to-end learning method, our method breaks the vision-based lateral control system down into a perception module and a control module. The perception module which is based on a multi-task learning neural network first takes a driver-view image as its input and predicts the track features. The control module which is based on reinforcement learning then makes a control decision based on these features. In order to improve the data efficiency, we propose visual TORCS (VTORCS), a deep reinforcement learning environment which is based on the open racing car simulator (TORCS).
In this paper, a policy iteration-based Q-learning algorithm is proposed to solve infinite horizon linear nonzero-sum quadratic differential games with completely unknown dynamics. The Q-learning algorithm, which employs off-policy reinforcement learning (RL), can learn the Nash equilibrium and the corresponding value functions online, using the data sets generated by behavior policies. First, we prove equivalence between the proposed off-policy Q-learning algorithm and an offline PI algorithm by selecting specific initially admissible polices that can be learned online. Then, the convergence of the off-policy Q-learning algorithm is proved under a mild rank condition that can be easily met by injecting appropriate probing noises into behavior policies. The generated data sets can be repeatedly used during the learning process, which is computationally effective. The simulation results demonstrate the effectiveness of the proposed Q-learning algorithm.
In this paper, the main issues of model-based control methods are first reviewed, and then followed by the motivations and the state-of-the-art of the model-free adaptive control (MFAC). MFAC is a novel data-driven control method for a class of unknown non-affine nonlinear discrete-time systems since neither explicit physical model nor Lyapunov stability theory or key technical lemma is used in the controller design and theoretical analysis except for the input/output (I/O) measurement data only. The basis of MFAC is the dynamic linearization data modeling method at each operating point of the closed-loop system. The established dynamic linearization data model is a virtual equivalent data relationship in the I/O sense to the original nonlinear plant by means of a novel concept called pseudopartial derivative (PPD) or pseudo-gradient (PG) vector. Based on this virtual equivalent dynamic linearization data model and the time-varying PPD or PG estimation algorithm designed merely using the I/O measurements of a controlled plant, the MFAC system is then constructed. The main contribution of this work is that the theoretical analysis of the bounded-input boundedoutput stability, the monotonic convergence of the tracking error dynamics, and the internal stability of the Full Form dynamic linearization based MFAC scheme, are rigorously presented by the contraction mapping principle, and the well known PID control and the traditional adaptive control for linear timeinvariant systems are explicitly shown as the special cases of this MFAC. The simulation results verify the effectiveness of the proposed approach.
Model predictive control (MPC)-based path following schemes for autonomous cars represent a novel and highly debated control approach. Their high online computational load poses a challenge for practical real-time application in vehicle systems with fast dynamics. This paper proposes an implementation scheme for an MPC path following controller that considers the feasible road region and vehicle shape. Moreover, the model mismatch induced by varying road conditions and small-angle assumptions is considered in the form of a measurable disturbance. To solve the optimization problem for the proposed MPC path following controller, a differential evolution (DE) algorithm is adopted. To verify the computational performance of the proposed implementation scheme, an experimental platform was developed that consists of the Hongqi autonomous car HQ430, various sensors, and systems for communication and data processing. The experimental results indicate that the proposed DE-based implementation strategy for the MPC path following controller achieves good computational performance and satisfactory control performance for path following in autonomous cars.
Feasible sets play an important role in model predictive control (MPC) optimal control problems (OCPs). This paper proposes a multi-parametric programming-based algorithm to compute the feasible set for OCP derived from MPC-based algorithms involving both spectrahedron (represented by linear matrix inequalities) and polyhedral (represented by a set of inequalities) constraints. According to the geometrical meaning of the inner product of vectors, the maximum length of the projection vector from the feasible set to a unit spherical coordinates vector is computed and the optimal solution has been proved to be one of the vertices of the feasible set. After computing the vertices, the convex hull of these vertices is determined which equals the feasible set. The simulation results show that the proposed method is especially efficient for low dimensional feasible set computation and avoids the non-unicity problem of optimizers as well as the memory consumption problem that encountered by projection algorithms.
This paper investigates the φ-type stability and robust stability for a general class of stochastic reaction-diffusion neural networks (SRDNNs) with Dirichlet boundary conditions, infinite discrete time-varying delays, and infinite continuously distributed delays. By virtue of inequality techniques, properties of M-matrix, and theories of stochastic analysis, several sufficient criteria are obtained to guarantee the almost sure φ-type stability, pth moment φ-type stability, and φ-type robust stability of the underlying SRDNNs with hybrid unbounded time delays. With appropriate choices of the function φ, the φ-type stability reduces to the exponential stability, polynomial stability, and logarithmic stability. Additionally, the developed results herein include some existing ones as special cases. A numerical simulation is performed to substantiate the effectiveness and superiority of the theoretical analysis.
The problem of output tracking is considered for Buck DC-DC converter system based on adaptive dynamic programming (ADP) algorithms. Firstly, an error system is constructed. Then, a performance index is defined by taking the tracking error and control cost into consideration. Next, the optimal control is obtained by ADP for the Buck converter system, which can achieve the optimal tracking error and control cost. The problem of output tracking is also studied for networked Buck converter system based on ADP. The newly proposed algorithms avoid solving Riccati equation algebraically and have the characteristics such as simplicity and quickness. Finally, simulation results are given to illustrate the validity of the proposed methods. © 2017, Editorial Department of Control Theory & Applications South China University of Technology. All right reserved.
In this paper, a new learning algorithm named OEM-ELM (Online Error Minimized-ELM) is proposed based on ELM (Extreme Learning Machine) neural network algorithm and the spreading of its main structure. The core idea of this OEM-ELM algorithm is: online learning, evaluation of network performance, and increasing of the number of hidden nodes. It combines the advantages of OS-ELM and EM-ELM, which can improve the capability of identification and avoid the redundancy of networks. The adaptive control based on the proposed algorithm OEM-ELM is set up which has stronger adaptive capability to the change of environment. The adaptive control of chemical process Continuous Stirred Tank Reactor (CSTR) is also given for application. The simulation results show that the proposed algorithm with respect to the traditional ELM algorithm can avoid network redundancy and improve the control performance greatly.
In this paper, a novel data-driven model-free adaptive predictive control method based on lazy learning technique is proposed for a class of discrete-time single-input and single-output nonlinear systems. The feature of the proposed approach is that the controller is designed only using the input-output (I/O) measurement data of the system by means of a novel dynamic linearization technique with a new concept termed pseudogradient (PG). Moreover, the predictive function is implemented in the controller using a lazy-learning (LL)-based PG predictive algorithm, such that the controller not only shows good robustness but also can realize the effect of model-free adaptive prediction for the sudden change of the desired signal. Further, since the LL technique has the characteristic of database queries, both the online and offline I/O measurement data are fully and simultaneously utilized to real-time adjust the controller parameters during the control process. Moreover, the stability of the proposed method is guaranteed by rigorous mathematical analysis. Meanwhile, the numerical simulations and the laboratory experiments implemented on a practical three-tank water level control system both verify the effectiveness of the proposed approach.
This paper presents a tracking control method for the lateral motion of an autonomous land vehicle (ALV). This method is based on active disturbance rejection control (ADRC) scheme and differential flatness theory. The lateral motion is hard to control since it is underactuated, nonlinear, and with large uncertainties. By making a small-angle approximation, the dynamic model is linearized. The flatness of the linear model is proved and a flat output is found. An equivalent form of the model is obtained based on the flat output and its derivatives only. Moreover, an ADRC is adopted to guarantee both control accuracy and strong robustness. Simulation results are presented and the results show the effectiveness of the control strategies.
This paper is concerned with the issues of feasibility, stability and robustness on the switched model predictive control (MPC) of a class of discrete-time switched linear systems with mode-dependent dwell time (MDT). The concept of conventional MDT in the literature of switched systems is extended to the stage MDT of lengths that vary with the stages of the switching. By computing the steps over which all the reachable sets of a starting region are contained into a targeting region, the minimum admissible MDT is offline determined so as to guarantee the persistent feasibility of MPC design. Then, conditions stronger than the criteria for persistent feasibility are explored to ensure the asymptotic stability. A concept of the extended controllable set is further proposed, by which the complete feasible region for given constant MDT can be determined such that the switched MPC law can be persistently solved and the resulting closed-loop system is asymptotically stable. The techniques developed for nominal systems lay a foundation for the same issues on systems with bounded additive disturbance, and the switched tube-based MPC methodology is established. A required “switched” tube in the form of mode-dependent cross section is determined by computing a mode-dependent generalized robust positive invariant set for each error subsystem between nominal subsystem and disturbed subsystem. The theoretical results are testified via an illustrative example of a population ecological system.
Online learning algorithms have been preferred in many applications due to their ability to learn by the sequentially arriving data. One of the effective algorithms recently proposed for training single hidden-layer feedforward neural networks (SLFNs) is online sequential extreme learning machine (OS-ELM), which can learn data one-by-one or chunk-by-chunk at fixed or varying sizes. It is based on the ideas of extreme learning machine (ELM), in which the input weights and hidden layer biases are randomly chosen and then the output weights are determined by the pseudo-inverse operation. The learning speed of this algorithm is extremely high. However, it is not good to yield generalization models for noisy data and is difficult to initialize parameters in order to avoid singular and ill-posed problems. In this paper, we propose an improvement of OS-ELM based on the bi-objective optimization approach. It tries to minimize the empirical error and obtain small norm of network weight vector. Singular and ill-posed problems can be overcome by using the Tikhonov regularization. This approach is also able to learn data one-by-one or chunk-by-chunk. Experimental results show the better generalization performance of the proposed approach on benchmark datasets.
In this paper, we develop an online sequential learning algorithm for single hidden layer feedforward networks (SLFNs) with additive or radial basis function (RBF) hidden nodes in a unified framework. The algorithm is referred to as online sequential extreme learning machine (OS-ELM) and can learn data one-by-one or chunk-by-chunk (a block of data) with fixed or varying chunk size. The activation functions for additive nodes in OS-ELM can be any bounded nonconstant piecewise continuous functions and the activation functions for RBF nodes can be any integrable piecewise continuous functions. In OS-ELM, the parameters of hidden nodes (the input weights and biases of additive nodes or the centers and impact factors of RBF nodes) are randomly selected and the output weights are analytically determined based on the sequentially arriving data. The algorithm uses the ideas of ELM of Huang et al. developed for batch learning which has been shown to be extremely fast with generalization performance better than other batch training methods. Apart from selecting the number of hidden nodes, no other control parameters have to be manually chosen. Detailed performance comparison of OS-ELM is done with other popular sequential learning algorithms on benchmark problems drawn from the regression, classification and time series prediction areas. The results show that the OS-ELM is faster than the other sequential algorithms and produces better generalization performance.
In this paper, a model predictive control (MPC) approach for controlling an active front steering system in an autonomous vehicle is presented. At each time step, a trajectory is assumed to be known over a finite horizon, and an MPC controller computes the front steering angle in order to follow the trajectory on slippery roads at the highest possible entry speed. We present two approaches with different computational complexities. In the first approach, we formulate the MPC problem by using a nonlinear vehicle model. The second approach is based on successive online linearization of the vehicle model. Discussions on computational complexity and performance of the two schemes are presented. The effectiveness of the proposed MPC formulation is demonstrated by simulation and experimental tests up to 21 m/s on icy roads
No Parametric Model and Adaptive Control Theory
- Z Hou
Model Predictive Control of Autonomous Driving Vehicles
- J Hong
- Y Jiang
The heterogeneous systems integration design and implementation for lane keeping on a vehicle
- S Wu
- H Chiang
- C Perng
- B Chen
- T T Wu
- Liu
Adaptive dynamic programming algorithms for the output tracking of buck converter systems
- J Jian
- Y Shen
- Y Liu