This study proposes a novel distributed predictive cruise control (PCC) algorithm based on reinforcement learning. The algorithm aims at reducing idle time and maintaining an adjustable speed depending on the traffic signals. The effectiveness of the proposed approach has been validated through Paramics microscopic traffic simulations by proposing a scenario in Statesboro, Georgia. For different traffic demands, the travel time and fuel consumption rate of vehicles are compared between non-PCC and PCC algorithms. Microscopic traffic simulation results demonstrate that the proposed PCC algorithm will reduce the fuel consumption rate by 4.24% and decrease the average travel time by 3.78%.

A fast-running human detection system for the unmanned aerial vehicle (UAV) based on optical flow and deep convolution networks is proposed in this study. In the system, running humans can be detected in real-time at the speed of 15 frames per second (fps) with an 81.1% detection accuracy. To fast locate the candidate targets, optical flow representing the motion information is calculated with every two successive frames. A series of prior-processing operations, including spatial average filtering, morphological expansion and outer contour extraction, are performed to extract the regions of interest. A classification model based on small-kernel convolution networks is proposed to achieve the accurate recognition of the running people in various backgrounds. In the model, small convolutional filters are adopted to accelerate the speed of the data representation. Moreover, a total of 60,000 samples are collected to enhance the robustness of the model to adapt to the complex outdoor UAV scenes. The proposed method is compared with other deep learning frameworks for object detection. Field experiments on UAV videos are performed to verify that the proposed system can effectively detect the running people targets in real-time.

Multi-hop routing in vehicular ad-hoc networks (VANETs) and wireless sensor networks has attracted significant interest of researchers in the wireless ad-hoc networks community. Most multi-hop routing protocols in VANET are based around the idea of choosing the next destination, which will provide the shortest-delay to reach a destination. To ensure better monitoring and reporting of road condition information, this study proposes location-based data forwarding through roadside sensors using k-shortest path routing combined with Q-learning. Q-learning is used for exploration of the sensing field to determine those sensors which have a higher queuing delay during peak hours as well as those which have comparatively lower delays. The use of Q-learning for exploration (sans routing) enables faster convergence for the sensors as compared to those techniques which utilise naive Q-learning for shortest path routing. Secondly, multi-hop routing is being combined with source coding (Huffman and Arithmetic coding) to compress the data payload of packets. This has shown some promising results for the VANETs employing dedicated short-range communication.

Decision making for self-driving cars is usually tackled by manually encoding rules from drivers’ behaviours or imitating drivers’ manipulation using supervised learning techniques. Both of them rely on mass driving data to cover all possible driving scenarios. This study presents a hierarchical reinforcement learning method for decision making of self-driving cars, which does not depend on a large amount of labelled driving data. This method comprehensively considers both high-level manoeuvre selection and low-level motion control in both lateral and longitudinal directions. The authors firstly decompose the driving tasks into three manoeuvres, including driving in lane, right lane change and left lane change, and learn the sub-policy for each manoeuvre. Then, a master policy is learned to choose the manoeuvre policy to be executed in the current state. All policies, including master policy and manoeuvre policies, are represented by fully-connected neural networks and trained by using asynchronous parallel reinforcement learners, which builds a mapping from the sensory outputs to driving decisions. Different state spaces and reward functions are designed for each manoeuvre. They apply this method to a highway driving scenario, which demonstrates that it can realise smooth and safe decision making for self-driving cars.

Abnormal driving behaviours, such as rapid acceleration, emergency braking, and rapid lane changing, bring great uncertainty to traffic, and can easily lead to traffic accidents. The accurate identification of abnormal driving behaviour helps to judge the driver's driving style, inform surrounding vehicles, and ensure the road traffic safety. Most of the existing studies use clustering and shallow learning, it is difficult to accurately identify the types of abnormal driving behaviours. Aimed at addressing the difficulty of identifying driving behaviour, this study proposed a recognition model based on a long short-term memory network and convolutional neural network (LSTM-CNN). The extreme acceleration and deceleration points are detected through the statistical analysis of real vehicle driving data, and the driving behaviour recognition data set is established. By using the data set to train the model, the LSTM-CNN can achieve a better result.

The use of deep learning methods to predict traffic flow in transportation systems has become a hot research project. The existing predictive model method faces problems such as long calculation time and difficult data pre-processing, especially for the prediction effect of high traffic area. In this study, the authors propose a novel framework ST-ESNet, spatio-temporal expand-and-squeeze networks, that designs several effective strategies for considering the complexity, non-linearity and uncertainty of traffic flow, and better captures traffic flow characteristics to adapt to the dynamic characteristics of traffic trajectory, traffic duration and traffic flow. Specially, we use extend-and-squeeze process rather than squeeze-and-extend process during the normal residual unit to capture farther spatial dependence among regions. Specifically, inverted residual and deformed convolution structures are utilised in the expanding process, and the convolution with stride 2 is utilised in the squeeze process. Furthermore, image feature scaling is used in each residual unit to obtain more fine-grained surface information, which improves the ability of the model to capture dynamic spatial dependence features. Finally, they use stochastic weight averaging to obtain an integration model. In summary, they propose a new predictive model ST-ESNet. The experimental results show that the authors’ proposed network model has better prediction performance compared with the state-of-the-art model.

The legibility of traffic signs has been considered from the beginning of design, and traffic signs are easy to identify for humans. For computer systems, however, identifying traffic signs still poses a challenging problem. Both image-processing and machine-learning algorithms are constantly improving, aimed at better solving this problem. However, with a dramatic increase in the number of traffic signs, labelling a large amount of training data means high cost. Therefore, how to use a small number of labelled traffic sign data reasonably to build an efficient and high-quality traffic sign recognition (TSR) model in the Internet-of-things–based (IOT-based) transport system has been an urgent research goal. Here, the authors propose a novel semi-supervised learning approach combining global and local features for TSR in an IOT-based transport system. In their approach, histograms of oriented gradient, colour histograms (CH), and edge features (EF) are used to build different feature spaces. Meanwhile, on the unlabelled samples, a fusion feature space is found to alleviate the differences between different feature spaces. Extensive evaluations on a collection of signs from the German Traffic Sign Recognition Benchmark (GTSRB) dataset shows that the proposed approach outperforms the others and provides a potential solution for practical applications.

Autonomous driving has been achieving great progress since last several years. However, the autonomous vehicles always ignore the important traffic information on the road because of the uncertainties of driving environment and the limitations of onboard sensors. This might cause serious safety problem in autonomous driving. This study argues that the connected vehicles could share much more environmental information with each other. Therefore, a decision-making method based on reinforcement learning is proposed for V2X autonomous vehicles. First, the V2X autonomous driving architecture with three subsystems is designed. By V2V communication, an autonomous vehicle could obtain much more environmental information. Second, a reinforcement learning based model is applied to learn from the V2V observation data. A simulation environment is setup based on OpenAI reinforcement learning framework. The experimental results demonstrate the effectiveness of the V2X in autonomous driving.

In autonomous driving, stereo vision-based depth estimation technology can help to estimate the distance of obstacles accurately, which is crucial for correctly planning the path of the vehicle. Recent work has formulated the stereo depth estimation problem into a deep learning model with convolutional neural networks. However, these methods need a lot of post-processing and do not have strong adaptive capabilities to ill-posed regions or new scenes. In addition, due to the difficulty of the labelling the ground truth depth for real circumstance, training data for the system is limited. To overcome the above problems, the authors came up with self-improving pyramid stereo network, which can not only get a direct regression disparity without complicated post-processing but also be robust in ill-posed area. Moreover, by online learning, the proposed model can not only address the data limitation problem but also save the time spent on training and hardware resources in practice. At the same time, the proposed model has a self-improving ability to new scenes, which can quickly adjust the model according to the test data in time and improve the accuracy of prediction. Experiments on Scene Flow and KITTI data set demonstrate the effectiveness of the proposed network.

This study considers an adaptive cruise control problem of connected vehicles in the vehicular ad-hoc network and proposes a Gaussian learning-based fuzzy predictive cruise control approach to enhance the fuel efficiency and safety of the connected vehicles in a vehicle-following scenario. First, a Gaussian process regression model is introduced and trained with real data to estimate the future acceleration of the preceding vehicle over the prediction horizon. Moreover, with assessing traffic scenarios, the weights characterising the importance of individual performance are adjusted by a fuzzy decision method in real time. Then a fuzzy predictive cruise controller is obtained by online solving a constrained receding horizon optimal control problem with a changing cost function and acceleration prediction of the preceding vehicle. Finally, through CarSim/Simulink co-simulation, it is shown that the proposed approach has an improvement in fuel economy and safety compared with conventional predictive cruise control algorithms.

The unmanned surface vessel (USV) plays an important role in smart ocean. This study proposes an adaptive fault-tolerant tracking control for USVs in the presence of input saturations and error constraints. A tan-type barrier Lyapunov function is utilised for the error constraints and the neural networks are employed to treat the model uncertainty. Moreover, the adaptive technique combined with the backstepping method not only enables the actuator fault-tolerant controller to address the fault effects but also handles the external disturbances and input saturations. The proposed control approach can track the desired trajectory with error constraints and the system is guaranteed to be uniformly bounded under certain actuator failure. Numerical simulation is carried out to verify the effectiveness of this control strategy.

The three-dimensional trajectory tracking of an underactuated autonomous underwater vehicle (AUV) under the complex unknowns including model uncertainties and time-varying disturbances is studied. The reference pitch angle and yaw angle are designed in kinematics, based on the time-varying reference trajectory. Dynamics controllers are developed by incorporating the first-order filter into the dynamic surface control (DSC), which simplifies the design process and overcome differential explosion in the traditional backstepping. To reduce the influence of the complex unknowns, an adaptive fuzzy-based DSC scheme is employed to identify the lumped disturbances with arbitrary accuracy and further enhance system robustness. Lyapunov stability analysis demonstrates that tracking errors are bounded and converge to an arbitrarily small neighbourhood of zero. Finally, simulation studies and comparisons with DSC scheme are carried out to illustrate the effectiveness and superiority of the proposed scheme.

In this study, an adaptive fuzzy inverse optimal control problem is investigated for a class of vehicle active suspension systems. Since active suspension systems have dynamic characteristics of complexities and spring non-linearities, the fuzzy logic systems are utilised to learn the unknown non-linear dynamics. In addition, there exist the constraints of the displacements of the sprung and unsprung masses, vertical vibration speeds, and current intensity in the considered suspension system, therefore, the Barrier Lyapunov functions are introduced into the control design to ensure that the full-state constraints are not overstepped. The inverse optimal control method is adopted by constructing an auxiliary system, which circumvents the assignment of solving a Hamilton–Jacobi–Bellman equation and brings about an inverse optimal controller associated with a meaningful objective functional. Based on Lyapunov stability theory and backstepping recursive design algorithm, a fuzzy adaptive optimal control scheme is developed. It is proved that the proposed control scheme not only guarantees that the vertical vibration of the vehicle is stabilised by the electromagnetic actuator but also achieves the goal of inverse optimality with regard to the cost functional. Finally, the simulation studies check the validity of the presented control strategy.

The performance of the torque control and distribution is a critical problem that can affect the stability, safety and energy efficiency of a rear-wheel independent drive (RWID) electric vehicle (EV). This study proposes a model adaptive torque control and distribution method for RWID EVs. First, the torque control and distribution problem is analysed in detail. Then an RWID EVs’ longitudinal model is built and a torque control and distribution scheme is proposed. To avoid the over-actuation and the under-actuation of the powertrain system, a controller is designed based on the longitudinal model to adaptively control the driving torque. To comprehensively consider the stability and safety, an error reconstruction strategy based on the fuzzy logic theory is proposed to evaluate the errors in the side slip angle and in the yaw rate. In order to accurately distribute the driving torque to the rear wheels, a torque distribution controller is designed. Finally, the proposed method is validated on a co-simulation platform, and the simulation results demonstrate the excellent performance of the proposed method for RWID EVs’ torque control and distribution compared with the counterparts of fuzzy logic direct yaw-moment control and two-loop torque distribution and control.

Pedestrian motion recognition is one of the important components of an intelligent transportation system. Since commonly used spatial-temporal features are still not sufficient for mining deep information in frames, this study proposes a three-stream neural network called a spatial-temporal-relational network (STRN), where the static spatial information, dynamic motion and differences between adjunct keyframes are comprehensively considered as features of the video records. In addition, an optimised pooling layer called convolutional vector of locally aggregated descriptors layer (Conv-VLAD) is employed before the final classification step in each stream to better aggregate the extracted features and reduce the inter-class differences. To accomplish this, the original video records are required to be processed into RGB images, optical flow images and RGB difference images to deliver the respective information for each stream. After the classification result is obtained from each stream, a decision-level fusion mechanism is introduced to improve the network's overall accuracy via combining the partial understandings together. Experimental results on two public data sets UCF101 (94.7%) and HMDB51 (69.0%), show that the proposed method achieves significantly improved performance. The results of STRN have far-reaching significance for the application of deep learning in intelligent transportation systems to ensure pedestrian safety.

To solve the unacceptable issue caused by the inconsistency of lane-changing behaviour between autonomous vehicles and actual drivers. A lane-changing behaviour decision-making model based on the Gauss mixture hidden Markov model (GM-HMM) is proposed according to the characteristic of a driver's lane changing behaviour. The proposed model is tested and verified based on the database of Next-Generation Simulation (NGSIM). The results show that the GM-HMM is 95.4% similar to the real driver's behaviour. To further verify the proposed model, the proposed algorithm is compared with some machine learning techniques from literature in different test scenarios. The comparison and analysis indicate that the GM-HMM method can more accurately simulate the real driver's lane-change behaviour, thus improving the trust of the passengers and other vehicles around autonomous vehicles.

Accurate knowledge of vehicle inertial parameters (e.g. vehicle mass and yaw moment of inertia) is essential to manage vehicle potential trajectories and improve vehicle active safety. For lightweight electric vehicles (LEVs), whose control performance of dynamics system can be substantially affected due to the drastic reduction of vehicle weights and body size, such knowledge is even more critical. This study proposes a dual unscented Kalman filter (DUKF) approach, where two UKFs run in parallel to simultaneously estimate vehicle states and parameters such as vehicle velocity, vehicle sideslip angle, and inertial parameters. The proposed method only utilises real-time measurements from torque information of in-wheel motor and sensors in a standard car. The four-wheel non-linear vehicle dynamics model considering payload variations is developed, local observability of the DUKF observer is analysed and derived via differential geometry theory. To address the non-linearities in vehicle dynamics, the DUKF and dual extended Kalman filter (DEKF) are also presented and compared. Simulations with various manoeuvres are carried out using the platform of MATLAB/Simulink-Carsim^®. Simulation results of MATLAB/Simulink-Carsim^® show that the proposed DUKF method can effectively estimate inertial parameters of LEV under different payloads. Moreover, the investigation reveals that the proposed DUKF approach has better performance of estimating vehicle inertial parameters compared with the DEKF method.

This study proposes a method for a vehicle controller to learn human driving behaviours through iterative interactions. In particular, the vehicle controller and the human driver jointly control a vehicle along a path only known to the human driver. Through repeated cooperative driving, the vehicle controller estimates the hidden desired path of the driver by minimising the control input. Eventually, semi-autonomous driving is realised since the vehicle controller is able to automatically track the target path and release the human driver from the driving task. The iterative learning of the human target path on the basis of the proposed algorithm is in the spatial domain, and is effective in the presence of uncertain human driving speeds. The validity of the proposed method is proved by rigorous analysis and demonstrated by numerical simulations.

Increasing electric vehicles (EV) charging efficiency and reducing charging cost are important issues of EVs power management and key factor that affect the spread of EVs. Researching the competitive relationship between power grids and EV users is the key to solving this problem. Therefore, the authors propose a charging control strategy that is based on a 1-N-type Stackelberg game. In the game, power grid, retailers, and users are all able to do power decision making. Thus, the charging strategy here can flexibly meet different demands of power grid, retailers, and users. The equilibrium of the game model is solved by the inverse induction method. The benefits are compared by simulation, by the disordered charging process, and by charging control methods based on static time-sharing electricity prices. The influence of the parameters on the charging process in the game model is analysed, and the feasibility of the proposed method is verified. Results show that load peak is reduced by 34.9% while users' charging costs are reduced by 25.1%. The approaches they proposed can effectively solve the competition between the two.

This study addresses the finite-time tracking control problem for autonomous airships with uncertainties and external disturbances. Six-degree-of-freedom kinematics and dynamics equations are established by considering the model uncertainties and external disturbances. To handle the model uncertainties and external disturbances, a finite-time sliding mode disturbance observer (DOB) is designed. To have a good tracking performance, a finite-time command-filtered backstepping-supertwisting controller is proposed with DOB. By using Lyapunov theory, the finite-time convergence of tracking errors and the stability of the control method is proved. Finally, the simulation results show that the controller can track the desired trajectory well in spite of model uncertainties and external disturbances.

This study presents a finite-time spatial path following control method for a robotic underactuated airship subject to model uncertainties and external disturbances. A finite-time path following approach is proposed by combining the backstepping approach with the terminal sliding mode control technique, and the upper bounds of model uncertainties and external disturbances are estimated by the designed adaptive laws. Compared with existing works on the path following control of airships, the algorithm presented in this study can guarantee the airship track a spatial predefined path in finite time in the presence of model uncertainties and external disturbances. Simulations are given to illustrate the effectiveness of the proposed path following control method.

This study is concerned with the sensor fault estimation problem for network-based vehicle suspension system with deny-of-service attack, where a linear robust observer is designed. First of all, the attack behaviour switching is modelled as a Markovian jumping process, and then a sufficient condition based on the Markovian jumping system approach is proposed such that the sensor fault estimation error system is asymptotically stable in the mean-square sense with a prescribed performance level. In this work, the occurring and transition probabilities of the attack are allowed to be partially unknown and uncertain. Finally, a simulation example is presented that validates the effectiveness of design method.

Vehicle mass is one of the most critical parameters in the vehicle control system, based on the discrete vehicle longitudinal dynamic equation after the forward Euler approximation, non-linear particle filter is introduced to estimate the vehicle mass intelligently, and it gains a competitive advantage that statistical characteristics of noise and uncertainties in the system are not necessary to be known or supposed in advance. As a sort of recursive, Bayesian state estimator, vehicle mass is regarded as a constant state variable to constitute the discrete state-space equation, motor torque is selected as input signal, and the measurable vehicle speed is selected to constitute the observation equation, parameters such as rolling resistance coefficient, air drag coefficient and road slop are considered as high-power noise and uncertainties. The performance of the proposed vehicle mass estimator is tested by several groups of load and the results demonstrate that the output of the particle filter based vehicle mass estimator can converge to the real value and keep steady.