Search

Self-driving cars are on the horizon, making it necessary to consider safety assurance and homologation of these autonomously operating vehicles. In this study, we systematically review literature ...

Background: Motion sickness is common within most forms of transport; it affects most of the population who experience varied symptoms at some stage in their lives. Thus far, there has been no specific method to quantify the predicted levels of motion sickness for a given vehicle design, task and route. Objective: To develop a motion sickness virtual prediction tool that includes the following inputs: human motion, vision, vehicle motion, occupant task and vehicle design. Method: A time domain analysis using a multi-body systems approach has been developed to provide the raw data for post-processing of vehicle motion, occupant motion and vision, based on a virtual route designed to provoke motion sickness, while the digital occupant undertakes a specific non-driving related task. Results: Predicted motion sickness levels are shared for a simple positional sweep of a vehicle cabin due to a prescribed motion and task. Two additional examples are shared within this study; first, it was found that the model can predict the difference found between sitting forwards and backwards in an autonomous vehicle. Second, analysis of a respected and independent study into auxiliary display height shows that the model can predict both relative and absolute levels between the two display heights congruent to the original physical experiment. Conclusion: It has been shown that the tool has been successful in predicting motion sickness in autonomous vehicles and is therefore of great use in guiding new future mobility solutions in the ability to tune vehicle dynamics and control alongside vision and design attributes.

Ride comfort can heavily influence user experience and therefore comprises one of the most important vehicle design targets. Although ride comfort has been heavily researched, there is still no definite solution to its accurate estimation. This can be attributed, to a large extent, to the subjective nature of the problem. Aim of this study was to explore the use of neural networks for the accurate estimation of ride comfort by combining anthropometric data and acceleration measurements. Different acceleration inputs, neural network architectures, training algorithms and objective functions were systematically investigated, and optimal parameters were derived. New insight into the influence of anthropometric data on ride comfort has been gained. The results indicate that the proposed method improves the accuracy of subjective ride comfort estimation compared to current standards. Neural networks were trained using data derived from a range of field trials involving ten participants, on public roads and controlled environment. A clustering and sensitivity analysis complements the study and identifies the most important factors influencing subjective ride comfort evaluation.

Automated vehicles (AV's) offer greater flexibility in cabin design particularly in a future where no physical driving controls are required. One common concept for an automated vehicle is to have both forward and rearward facing seats. However, traveling backwards could lead to an increased likelihood of experiencing motion sickness due to the inability of occupants to anticipate the future motion trajectory. This study aimed to empirically evaluate the impact of seating orientation on the levels of motion sickness within an AV cabin. To this end, a vehicle was modified to replicate the common concept of automated vehicles with forward and rearward facing seats. Two routes were chosen to simulate motorway and urban driving. The participants were instructed to carry out typical office tasks whilst being driven in the vehicle which consisted of conducting a meeting, operating a personal device and taking notes. The participants conducted the test twice to experience both forward and rearward seating orientations in a randomised crossover design. Levels of sickness reported was relatively low with a significant increase in the mean level of sickness recorded when traveling rearwards. As expected, this increase was particularly pronounced under urban driving conditions. It is concluded that rearward travel in automated vehicles will compromise the passenger experience.

With the development and deployment of lightweight vehicles to the market, inclusive of autonomous pods, a review of advanced crashworthy structures and the design methodology has been conducted as it is thought that super-lightweight vehicles may pose significant risk to the occupants if they are involved in a crash. It is suggested that tests should include oblique and multiple velocity impacts to cater for the effects of assisted driving systems of future vehicles. A review of current crash structures and design methodologies revealed that the most recent research do not cater to multiple crash scenarios, nor a shorter crush allowance, therefore resulting in poor crashworthiness performance. In addition, the arbitrary seat positioning shown in autonomous pods’ concepts vastly increases the risk to occupants. Greater enhancements to passive crashworthiness are imperative. To this end, functionally graded vehicle structures should be designed as it has been found that these can provide optimized solutions. Research into nonlinear optimization methods for computationally expensive problems will become central to this.

The high level of air pollution in urban areas, caused in no small extent by road transport, requires the implementation of continuous and accurate monitoring techniques if emissions are to be minimized. The primary motivation for this paper is to enable fine spatiotemporal monitoring based on crowd sensing, whereby the instantaneous fuel consumption of a vehicle is estimated using smartphone measurements. To this end, a surrogate method based on indirect monitoring using Recurrent Neural Networks (RNNs) that process a smartphone's GPS position, speed, altitude, acceleration and number of visible satellites is proposed. Extensive field trials were conducted to gather smartphone and fuel consumption data at a wide range of driving conditions. Two different RNN types were explored, and a parametric analysis was performed to define a suitable architecture. Various training methods for tuning the RNN were evaluated based on performance and computational burden. The resulting estimator was compared with others found in the literature, and the results confirm its superior performance. The potential impact of the proposed method is noteworthy as it can facilitate accurate monitoring of in-use vehicle fuel consumption and emissions at large scales by exploiting available smartphone measurements.

In this paper, a deep learning approach is proposed to accurately position wheeled vehicles in Global Navigation Satellite Systems (GNSS) deprived environments. In the absence of GNSS signals, information on the speed of the wheels of a vehicle (or other robots alike), recorded from the wheel encoder, can be used to provide continuous positioning information for the vehicle, through the integration of the vehicle's linear velocity to displacement. However, the displacement estimation from the wheel speed measurements are characterised by uncertainties, which could be manifested as wheel slips or/and changes to the tyre size or pressure, from wet and muddy road drives or tyres wearing out. As such, we exploit recent advances in deep learning to propose the Wheel Odometry neural Network (WhONet) to learn the uncertainties in the wheel speed measurements needed for correction and accurate positioning. The performance of the proposed WhONet is first evaluated on several challenging driving scenarios, such as on roundabouts, sharp cornering, hard-brake and wet roads (drifts). WhONet's performance is then further and extensively evaluated on longer-term GNSS outage scenarios of 30s, 60s, 120s and 180s duration, respectively over a total distance of 493 km. The experimental results obtained show that the proposed method is able to accurately position the vehicle with up to 93% reduction in the positioning error of its original counterpart after any 180s of travel. WhONet's implementation can be found at https://github.com/onyekpeu/WhONet.

Recurrent Neural Networks (RNNs) are known for their ability to learn relationships within temporal sequences. Gated Recurrent Unit (GRU) networks have found use in challenging time-dependent applications such as Natural Language Processing (NLP), financial analysis and sensor fusion due to their capability to cope with the vanishing gradient problem. GRUs are also known to be more computationally efficient than their variant, the Long Short-Term Memory neural network (LSTM), due to their less complex structure and as such, are more suitable for applications requiring more efficient management of computational resources. Many of such applications require a stronger mapping of their features to further enhance the prediction accuracy. A novel Quaternion Gated Recurrent Unit (QGRU) is proposed in this paper, which leverages the internal and external dependencies within the quaternion algebra to map correlations within and across multidimensional features. The QGRU can be used to efficiently capture the inter- and intra-dependencies within multidimensional features unlike the GRU, which only captures the dependencies within the sequence. Furthermore, the performance of the proposed method is evaluated on a sensor fusion problem involving navigation in Global Navigation Satellite System (GNSS) deprived environments as well as a human activity recognition problem. The results obtained show that the QGRU produces competitive results with almost 3.7 times fewer parameters compared to the GRU. The QGRU code is available at https://github.com/onyekpeu/Quarternion-Gated-Recurrent-Unit.

The safety of vulnerable road users is of paramount importance as transport moves towards fully automated driving. The richness of real-world data required for testing autonomous vehicles is limited and furthermore, available data do not present a fair representation of different scenarios and rare events. Before deploying autonomous vehicles publicly, their abilities must reach a safety threshold, not least with regards to vulnerable road users, such as pedestrians. In this paper, we present a novel Generative Adversarial Networks named the Ped-Cross GAN. Ped-Cross GAN is able to generate crossing sequences of pedestrians in the form of human pose sequences. The Ped-Cross GAN is trained with the Pedestrian Scenario dataset. The novel Pedestrian Scenario dataset, derived from existing datasets, enables training on richer pedestrian scenarios. We demonstrate an example of its use through training and testing the Ped-Cross GAN. The results show that the Ped-Cross GAN is able to generate new crossing scenarios that are of the same distribution from those contained in the Pedestrian Scenario dataset. Having a method with these capabilities is important for the future of transport, as it will allow for the adequate testing of Connected and Autonomous Vehicles on how they correctly perceive the intention of pedestrians crossing the street, ultimately leading to fewer pedestrian casualties on our roads.

Road localisation of autonomous vehicles is reliant on consistent accurate GNSS (Global Navigation Satellite System) positioning information. Commercial GNSS receivers usually sample at 1 Hz, which is not sufficient to robustly and accurately track a vehicle in certain scenarios, such as driving on the highway, where the vehicle could travel at medium to high speeds, or in safety-critical scenarios. In addition, the GNSS relies on a number of satellites to perform triangulation and may experience signal loss around tall buildings, bridges, tunnels and trees. An approach to overcoming this problem involves integrating the GNSS with a vehicle-mounted Inertial Navigation Sensor (INS) system to provide a continuous and more reliable high rate positioning solution. INSs are however plagued by unbounded exponential error drifts during the double integration of the acceleration to displacement. Several deep learning algorithms have been employed to learn the error drift for a better positioning prediction. We therefore investigate in this chapter the performance of Long Short-Term Memory (LSTM), Input Delay Neural Network (IDNN), Multi-Layer Neural Network (MLNN) and Kalman Filter (KF) for high data rate positioning. We show that Deep Neural Network-based solutions can exhibit better performances for high data rate positioning of vehicles in comparison to commonly used approaches like the Kalman filter.

Automated Vehicles and next generation ADAS hold the promise of disrupting mobility. However, public field trials have recently highlighted road anomalies, such as potholes and bumps, as a source of autopilot disengagements. In this paper, we research the influence of road anomalies on the performance of Artificial Intelligence-based vision systems. To this end, we conducted controlled real-world experiments and developed a validated vehicle system computational model using IPG Carmaker. The vehicle detection, tracking and distance estimation performance have been investigated by undertaking a thorough sensitivity analysis. The results indicate the system limitations in performing adequately for a range of bump sizes and vehicle speeds. With our findings we put emphasis on the importance of vehicle dynamics in the development of automated driving systems.

Chassis Active Safety Systems require access to a set of vehicle dynamics motion states which measurement is neither trivial nor cost-effective (e.g. lateral velocity). In this work, virtual sensing is applied to vehicle dynamics and proposed as a cost-effective solution to infer the vehicle planar motion states and three-axis tyre forces from signals measured by inexpensive sensors. Specifically, the tyre longitudinal forces are estimated using Adaptive Random-Walk Linear Kalman Filters and the vehicle planar motion states are determined in a hybrid state estimator formed by an Unscented Kalman Filter and Feedforward Neural Networks. The tyre vertical forces are estimated using a quasi-static weight transfer approach and Recursive Least Squares. The complete structure is integrated into a modular fashion and tested experimentally using a driver-in-the-loop setup. An extensive catalogue of manoeuvres is executed by a real driver to evidence the performance of the proposed virtual sensor at the limits of handling.

Vulnerable road user safety is of paramount importance as transport moves towards fully autonomous driving. The research question posed by this research is of how can we train a computer to be able to see and perceive a pedestrian’s movement. This work presents a dual network architecture, trained in tandem, which is capable of classifying the behaviour of a pedestrian from a single image with no prior context. The results show that the most successful network was able to achieve a correct classification accuracy of 94.3% when classifying images based on their behaviour. This shows the use of a novel data fusion method for pedestrian images and human poses. Having a network with these capabilities is important for the future of transport, as it will allow vehicles to correctly perceive the intention of pedestrians crossing the street, and will ultimately lead to fewer pedestrian casualties on our roads.

Nowadays, more than half of the world’s web traffic comes from mobile phones, and by 2020 approximately 70 percent of the world’s population will be using smartphones. The unprecedented market penetration of smartphones combined with the connectivity and embedded sensing capability of smartphones is an enabler for the large-scale deployment of Intelligent Transportation Systems (ITS). On the downside, smartphones have inherent limitations such as relatively limited energy capacity, processing power, and accuracy. These shortcomings may potentially limit their role as an integrated platform for monitoring driver behaviour in the context of ITS. This study examines this hypothesis by reviewing recent scientific contributions. The Cybernetics theoretical framework was employed to allow a systematic comparison. First, only a few studies consider the smartphone as an integrated platform. Second, a lack of consistency between the approaches and metrics used in the literature is noted. Last but not least, areas such as fusion of heterogeneous information sources, Deep Learning and sparse crowd-sensing are identified as relatively unexplored, and future research in these directions is suggested.

Safe and reliable operation of power plants invariably relies on the structural integrity assessments of pressure vessels and piping systems. Welded joints are a potential source of failure, because of the combination of the variation in mechanical properties and the residual stresses associated with the thermomechanical cycles experienced by the material during welding. This paper presents comparative studies between methods based on artificial neural networks (ANN) and fuzzy neural networks (FNN) for predicting residual stresses induced by welding. The performance of neural network and neuro-fuzzy systems are compared based on statistical indicators, scatter plots and several case studies. Results show that the neuro-fuzzy systems optimised using a hybrid technique can perform slightly better than a neural network trained using Levenberg-Marquardt algorithm, primarily because of the inability of the ANN approach to provide conservative estimates of residual stress profiles. Specifically, the prediction accuracy of the neuro-fuzzy systems trained using the hybrid technique is better for the axial residual stress component, with root mean square error (RMSE), absolute fraction of variance (R2) and mean absolute percentage error (MAPE) error of 0.1264, 0.9102 and 22.9442 respectively using the test data. Furthermore, this study demonstrates the potential benefits of implementing neuro-fuzzy systems in predicting residual stresses for use in structural integrity assessment of power plant components.

This paper presents a robust virtual sensor to estimate the chassis planar motion states and the tire forces during agile maneuvers using a tire-model-less approach. Specifically, virtual sensing is achieved from standard sensor signals available on the CAN bus of modern vehicles using a modular filter architecture composed of stochastic Kalman filters. A high-fidelity virtual testing environment is constructed in IPG CarMaker using a driver-in-the-loop setup to verify the virtual sensor without compromising driver's safety. Moreover, road random profiles are incorporated into the virtual road to assess the state estimator robustness to high vertical excitation levels. The virtual sensor is simulated under drifting maneuvers performed by an experienced test driver and tested experimentally under Fishhook and Slalom maneuvers. Finally, the state estimator is integrated into a drift controller, and autonomous drift control using exclusively readily available measurements is verified for the first time. As the drift equilibrium depends strongly on the tire–road friction, an adaptive neurofuzzy inference system has been integrated into the virtual sensor structure to provide a continuous approximation of the road friction characteristics (axle lateral force versus slip curve) in rigid and loose surfaces. The findings suggest that it may be possible to develop advanced vehicle controllers without using a tire model. This can lead to a substantial acceleration of development time, particularly in off-road applications, and remove the need for online estimation of tire properties due to pressure, wear, and age.

Topology optimisation is a process that is becoming increasingly reliable and necessary in the pursuit of highly efficient components comprising of low mass with a high structural performance. Thes...

This paper introduces the setup of the European network ITEAM aimed at the training of early-stage researchers (ESR) in the field of multi-actuated ground vehicles (MAGV). A network concept includes three main domains, where fifteen interconnected individual research projects are allocated: MAGV Integration, Green MAGV, and MAGV Driving Environment. All the projects are being carried out within the framework of continuous interdisciplinary training. The paper is focused on emerging research and technological trends, which are under elaboration in ITEAM projects, and the role of practice-oriented educational methods for their realization.

This paper examines the implementation of a standalone 1 MW Wave Energy Converter (WEC), based on a novel concept of a class of WECs, consisting in fully enclosed appropriate internal body configurations, which provide inertial reaction against the motion of an external vessel. Acting under the excitation of the waves, the external vessel is subjected to a simultaneous surge and pitch motion in all directions, ensuring maximum wave energy capture. The internal body is suspended from the external vessel body in such an appropriate geometrical configuration, that a symmetric four bar mechanism is essentially formed. The first advantage of this suspension geometry is that a linear trajectory results for the center of the mass of the suspended body with respect to the external vessel, enabling the introduction of a quite simple form of a Power Take-Off (PTO) design. The simplicity and symmetry of the suspension geometry and of the PTO, ensure a quite simple and robust technological implementation. Mass and inertia distribution of the internal body are optimized, ensuring maximal conversion and storage of wave energy. As a result, the internal body assembly is essentially, dynamically equivalent to a vertical physical pendulum. However, the resulting equivalent pendulum length and inertia can far exceed those that can be achieved by a simple horizontal or vertical pendulum, suspended or inverted, leading to a significant reduction of the suspended mass.

Design of a transferable time series anomaly detection method.Novel deep neural network structure facilitates learning short and long-term pattern interdependencies.Detection of anomalies in the Seismic Electrical Signal for predicting earthquake activity.Detection of road anomalies using smartphone data, facilitating crowdsourcing applications. The quest for more efficient real-time detection of anomalies in time series data is critically important in numerous applications and systems ranging from intelligent transportation, structural health monitoring, heart disease, and earthquake prediction. Although the range of application is wide, anomaly detection algorithms are usually domain specific and build on experts knowledge. Here a new signal processing algorithm inspired by the deep learning paradigm is presented that combines wavelets, neural networks, and Hilbert transform. The algorithm performs robustly and is transferable. The proposed neural network structure facilitates learning short and long-term pattern interdependencies; a task usually hard to accomplish using standard neural network training algorithms. The paper provides guidelines for selecting the neural network's buffer size, training algorithm, and anomaly detection features. The algorithm learns the system's normal behavior and does not require the existence of anomalous data for assessing its statistical significance. This is an essential attribute in applications that require customization. Anomalies are detected by analysing hierarchically the instantaneous frequency and amplitude of the residual signal using probabilistic Receiver Operating Characteristics. The method is shown to be able to automatically detect anomalies in the Seismic Electrical Signal that could be used to predict earthquake activity. Furthermore, the method can be used in combination with crowdsourcing of smartphone data to locate road defects such as potholes and bumps for intervention and repair.

The present study discusses the mechanical behaviour and modelling of a prototype automotive magnetorheological (MR) damper, which presents different viscous damping coefficients in jounce and rebound. The force generated by the MR damper is measured at different velocities and electrical currents, and a modified damper model is proposed to improve fitting of the experimental data. The model is calibrated by means of parameter identification, and for this purpose a new swarm intelligence algorithm is proposed, that we call the contrast-based Fruit Fly Optimisation Algorithm (c-FOA). The performance of c-FOA is compared with that of Genetic Algorithms, Particle Swarm Optimisation, Differential Evolution and Artificial Bee Colony. The comparison is made on the basis of no a-priori knowledge of the damper model parameters range. The results confirm the good performance of c-FOA under parametric range uncertainty. A sensitivity analysis discusses c-FOA’s performance with respect to its tuning parameters. Finally, a ride comfort simulation study quantifies the discrepancies in the results, for different identified damper model sets. The discrepancies underline the importance of accurately describing MR damper nonlinear behaviour, considering that virtual sign-off processes are increasingly gaining momentum in the automotive industry.

This paper proposes a new method of concurrent SOC and SOH estimation using a combination of recursive least square (RLS) algorithm and particle swarm optimization (PSO). The RLS algorithm is equipped with multiple fixed forgetting factors (MFFF) which are optimized by PSO. The performance of the hybrid RLS-PSO is compared with the similar RLS which is optimized by single objective genetic algorithms (SOGA) as well as multi-objectives genetic algorithm (MOGA). Open circuit voltage (OCV) is treated as a parameter to be estimated at the same timewith internal resistance. Urban Dynamometer Driving Schedule (UDDS) is used as the input data. Simulation results show that the hybrid RLS-PSO algorithm provides little better performance than the hybrid RLS-SOGA algorithm in terms of mean square error (MSE) and a number of iteration. On the other hand, MOGA provides Pareto front containing optimum solutions where a specific solution can be selected to have OCV MSE performance as good as PSO.

Truss optimization using fruit fly optimization algorithm.Advanced modelling of fruit fly food search behaviour.Efficiency in truss optimization with frequency constraints.Intuitive, few tuning parameters. A recent biological study shows that the extremely good efficiency of fruit flies in finding food, despite their small brain, emerges by two distinct stimuli: smell and visual contrast. contrast-based fruit fly optimization, presented in this paper, is for the first time mimicking this fruit fly behaviour and developing it as a means to efficiently address multi-parameter optimization problems. To assess its performance a study was carried out on ten mathematical and three truss optimization problems. The results are compared to those obtained using twelve state-of-the-art optimization algorithms and confirm its good and robust performance. A sensitivity analysis and an evaluation of its performance under parallel computing were conducted. The proposed algorithm has only a few tuning parameters, is intuitive, and multi-faceted, allowing application to complex n-dimensional design optimization problems.

The safety of vulnerable road users is of paramount importance as transport moves towards fully automated driving. The richness of real-world data required for testing autonomous vehicles is limited, and furthermore, the available data does not have a fair representation of different scenarios and rare events. This work presents a novel approach for the generation of human pose structures, specifically the type of pose structures that would appear to be in pedestrian scenarios. The results show that the generated pedestrian structures are indistinguishable from the ground truth pose structures when classified using a suitably trained classifier. The paper demonstrates that the Generative Adversarial Network architecture can be used to create realistic new training samples, and, in future, new pedestrian events.

A Vehicle with multiple drivetrains, like a hybrid electric one, is an over-actuated system that means there is an infinite number of combinations of torques that individual drivetrains can supply to provide a given total torque demand. Energy efficiency is considered as the secondary objective to determine the optimum solution among these feasible combinations. The resulting optimisation problem, which is nonlinear due to the multi-modal operation of electric machines, must be solved quickly to comply with the stability requirements of the vehicle dynamics. A theorem is developed for the first time to formulate and parametrically solve the energy-efficient torque distribution problem of a vehicle with multiple different drivetrains. The parametric solution is deployable on an ordinary electronic control unit (ECU) as a small-size lookup table that makes it significantly fast in operation. The fuel-economy of combustion engines, load transformations due to longitudinal and lateral accelerations, and traction efficiency of the off-road conditions are integrated into the developed theorem. Simulation results illustrate the effectiveness of the provided optimal strategy as torque distributors of on-road and off-road electrified vehicles with multiple different drivetrains.

Automated vehicles will provide greater transport convenience and interconnectivity, increase mobility options to young and elderly people, and reduce traffic congestion and emissions. However, the largest obstacle towards the deployment of automated vehicles on public roads is their safety evaluation and validation. Undeniably, the role of cameras and Artificial Intelligence-based (AI) vision is vital in the perception of the driving environment and road safety. Although a significant number of studies on the detection and tracking of vehicles have been conducted, none of them focused on the role of vertical vehicle dynamics. For the first time, this paper analyzes and discusses the influence of road anomalies and vehicle suspension on the performance of detecting and tracking driving objects. To this end, we conducted an extensive road field study and validated a computational tool for performing the assessment using simulations. A parametric study revealed the cases where AI-based vision underperforms and may significantly degrade the safety performance of AVs.

Recurrent neural networks (RNN) are distinguishable form other classes of artificial neural networks by their ability to make nodal connections along temporal sequences. Gated Recurrent Unit (GRU) proposed by Cho et al have found use in several time dependent applications such as natural language processing (NLP), financial analysis and sensor fusion applications due to their immunity to the vanishing gradient problem. GRU’s are also known to be more computationally efficient than their variant Long Short-term memory neural network (LSTM) due to their less complex structure and as such are more suitable for applications requiring more efficient management of computational resources. Many of such applications require a stronger mapping of their features to further enhance the prediction accuracy. A novel Quaternion gated recurrent unit (QGRU) is proposed which leverages the internal and external dependencies within the quaternion algebra to map correlations within and across multidimensional features. QGRU can be used to efficiently capture the inter and intra dependencies within multidimensional features unlike the GRU which only captures the dependencies within the sequence. Furthermore, the performance of the algorithm is evaluated on a sensor fusion problem involving navigation with INS sensors in GPS deprived environments.&nbsp

Precipitation can adversely influence road safety. Slippery road conditions have traditionally been detected using reactive methods requiring considerable excitation of the tire forces. Alternatives rely on non-contact methods such as vision, sound or ultrasonic sensors. This study proposes a cost-effective wet road conditions detection method based on acoustic measurements for urban and highway driving. It compared the performance of a range of machine learning algorithms to classify the road condition based on the audio features calculated using octave-band frequency analysis. The approach was evaluated experimentally using data collected from a vehicle instrumented with a microphone, GPS and CAN bus data logger. Support Vector Machines using Quadratic and Cubic kernels, as well as Logistic Regression performed better compared to other machine learning-based methods.

© Springer Nature Switzerland AG 2019. With the continued advancement of autonomous vehicles and their implementation in public roads, accurate detection of vulnerable road users (VRUs) is vital for ensuring safety. To provide higher levels of safety for these VRUs, an effective detection system should be employed that can correctly identify VRUs in all types of environments (e.g. VRU appearance, crowded scenes) and conditions (e.g. fog, rain, night-time). This paper presents optimal methods of sensor fusion for pedestrian and cyclist detection using Deep Neural Networks (DNNs) for higher levels of feature abstraction. Typically, visible sensors have been utilized for this purpose. Recently, thermal sensors system or combination of visual and thermal sensors have been employed for pedestrian detection with advanced detection algorithm. DNNs have provided promising results for improving the accuracy of pedestrian and cyclist detection. This is because they are able to extract features at higher levels than typical hand-crafted detectors. Previous studies have shown that amongst the several sensor fusion techniques that exist, Halfway Fusion has provided the best results in terms of accuracy and robustness. Although sensor fusion and DNN implementation have been used for pedestrian detection, there is considerably less research undertaken for cyclist detection.

Safety is an essential aspect in the facilitation of automated vehicle deployment. Current testing practices are not enough, and going beyond them leads to infeasible testing requirements, such as needing to drive billions of kilometres on public roads. Automated vehicles are exposed to an indefinite number of scenarios. Handling of the most challenging scenarios should be tested, which leads to the question of how such corner cases can be determined. We propose an approach to identify the performance boundary, where these corner cases are located, using Gaussian Process Classification. We also demonstrate the classification on an exemplary traffic jam approach scenario, showing that it is feasible and would lead to more efficient testing practices., Comment: 6 pages, 5 figures, accepted at 2019 IEEE Intelligent Transportation Systems Conference - ITSC 2019, Auckland, New Zealand, October 2019

Laser shock peening (LSP) as a surface treatment technique can improve the fatigue life and corrosion resistance of metallic materials by introducing significant compressive residual stresses near the surface. However, LSP-induced residual stresses are known to be dependent on a multitude of factors, such as laser process variables (spot size, pulse width and energy), component geometry, material properties and the peening sequence. In this study, an intelligent system based on machine learning was developed that can predict the residual stress distribution induced by LSP. The system can also be applied to “reverse-optimise” the process parameters. The prediction system was developed using residual stress data derived from incremental hole drilling. We used artificial neural networks (ANNs) within a Bayesian framework to develop a robust prediction model validated using a comprehensive set of case studies. We also studied the relative importance of the LSP process parameters using Garson’s algorithm and parametric studies to understand the response of the residual stresses in laser peening systems as a function of different process variables. Furthermore, this study critically evaluates the developed machine learning models while demonstrating the potential benefits of implementing an intelligent system in prediction and optimisation strategies of the laser shock peening process.

An approach based on Artificial Neural Networks is proposed in this paper to improve the localisation accuracy of Inertial Navigation Systems (INS)/Global Navigation Satellite System (GNSS) based aided navigation during the absence of GNSS signals. The INS can be used to continuously position autonomous vehicles during GNSS signal losses around urban canyons, bridges, tunnels and trees, however, it suffers from unbounded exponential error drifts cascaded over time during the multiple integrations of the accelerometer and gyroscope measurements to position. More so, the error drift is characterised by a pattern dependent on time. This paper proposes several efficient neural network-based solutions to estimate the error drifts using Recurrent Neural Networks, such as the Input Delay Neural Network (IDNN), Long Short-Term Memory (LSTM), Vanilla Recurrent Neural Network (vRNN), and Gated Recurrent Unit (GRU). In contrast to previous papers published in literature, which focused on travel routes that do not take complex driving scenarios into consideration, this paper investigates the performance of the proposed methods on challenging scenarios, such as hard brake, roundabouts, sharp cornering, successive left and right turns and quick changes in vehicular acceleration across numerous test sequences. The results obtained show that the Neural Network-based approaches are able to provide up to 89.55% improvement on the INS displacement estimation and 93.35% on the INS orientation rate estimation.

In the UK, the number of fatal accidents on rural roads is approximately double that on urban roads. Statistics have also shown that accidents on rural roads decreased less than on other road types. The narrow width and complex geometry are less forgiving to drivers’ mistakes. A potential remedy for this problem is automated driving in which the ability to plan -in real time- safe and feasible paths is essential. The literature review of recently proposed path planning methods has revealed that most of them utilise either forward simulations of a vehicle dynamics model or describe a priori mathematically a reference path. In this paper, the weaknesses of the reviewed methods are discussed and a new path planning method that belongs to the latter category is presented. The method is based on a direct element concept and as shown and discussed is extremely versatile. It is unique in the sense that for the first time it facilitates the prediction of the maximum vehicle slip angle and the definition of a reference path that minimises it. Contrary to other methods it is very flexible in defining arbitrary boundary and intermediate conditions. The overall computational cost as analysed is very small. Simulations illustrate its performance and comparisons with other methods highlight its strengths.

Topology optimisation is an increasingly important process used in a variety of industries to improve the designs of manufacturable products. The higher reliance of optimisation software, used for instance in the automotive industry, highlights its importance for designing more efficient and refined mass-produced components. Post-processing of topology optimisation results (e.g. from variable density to manufacturable structures) does however remain a heavily heuristic process where the end-results (and consequently the “efficiency” of the optimised product) can vary significantly as a function of the individual designer/engineer. This “variation” coupled with the often-significant time associated with post-processing makes the use of topology optimisation prohibitive in certain instances. In this paper, a systematic and repeatable three-step approach to automated post-processing of topology optimisation results for sheet metal manufacturing of automotive components will be introduced. The method, which has been implemented into a software tool, is mesh independent and can handle topology optimisation results in binary as well as variable density formats. The software contains three main steps; namely geometry refinement, re-analysis and manufacturability check. The methodology and software utilise a stencil method, for which the principles are described here. The main objective from this is to generate repeatable refined interpretations of optimisation results. In addition to presenting the actual methodology and software, this paper also investigates different parameter variations; such as geometry update sequence, search radii, stencil shape and type and their influence on the generated post-processed result. Definition of algorithm parameters is provided, together with suggested user-defined settings to enable the derivation of consistent refinements of the topology results.

This article presents a dataset recorded with a sensor-equipped research vehicle on public roads in the city of Coventry in the United Kingdom. The sensor suite includes a monocular-, infrared- and smartphone-camera, as well as a LiDAR unit, GPS receiver, smartphone sensors and vehicle CAN bus data logger. Data were collected by day and night in a variety of traffic, weather and road surface conditions with a focus on the correlation between vehicle dynamics and the environmental perception process of automated vehicles.

This paper presents an Optimized Unscented Kalman Filter for vehicle dynamics virtual sensing. An automated procedure to optimize the virtual sensor parameters based on metaheuristic algorithms is presented in order to avoid the time-consuming and complex manual tuning task. Specifically, Genetic Algorithm Optimization (GA) and contrast-based Fruit Fly optimization (c-FOA) are applied to minimize the estimation error in steady-state and transient driving maneuvers. The virtual sensor is implemented in a high-fidelity vehicle dynamics simulation software (IPG-CarMaker ®) and results demonstrate the improvement of the estimation accuracy with respect to a preliminary filter tuning carried out using a systematic trial and error approach.

The KDamper is a novel passive vibration isolation and damping concept, based essentially on the optimal combination of appropriate stiffness elements, which include a negative stiffness element. The KDamper concept does not require any reduction in the overall structural stiffness, thus overcoming the corresponding inherent disadvantage of the ‘‘Quazi Zero Stiffness’’ (QZS) isolators, which require a drastic reduction of the structure load bearing capacity. Compared to the traditional Tuned Mass damper (TMD), the KDamper can achieve better isolation characteristics, without the need of additional heavy masses, as in the case of the T Tuned Mass damper. Contrary to the TMD and its variants, the KDamper substitutes the necessary high inertial forces of the added mass by the stiffness force of the negative stiffness element. Among others, this can provide comparative advantages in the very low frequency range. The paper proceeds to a systematic analytical approach for the optimal design and selection of the parameters of the KDamper, following exactly the classical approach used for the design of the Tuned Mass damper. It is thus theoretically proven that the KDamper can inherently offer far better isolation and damping properties than the Tuned Mass damper. Moreover, since the isolation and damping properties of the KDamper essentially result from the stiffness elements of the system, further technological advantages can emerge, in terms of weight, complexity and reliability. A simple vertical vibration isolation example is provided, implemented by a set of optimally combined conventional linear springs. The system is designed so that the system presents an adequate static load bearing capacity, whereas the Transfer Function of the system is below unity in the entire frequency range. Further insight is provided to the physical behavior of the system, indicating a proper phase difference between the positive and the negative stiffness elastic forces. This fact ensures that an adequate level of elastic forces exists throughout the entire frequency range, able to counteract the inertial and the external excitation forces, whereas the damping forces and the inertia forces of the additional mass remain minimal in the entire frequency range, including the natural frequencies. It should be mentioned that the approach presented does not simply refer to discrete vibration absorption device, but it consists a general vibration absorption concept, applicable also for the design of advanced materials or complex structures. Such a concept thus presents the potential for numerous implementations in a large variety of technological applications, whereas further potential may emerge in a multi-physics environment. ispartof: Journal Of Vibration And Control vol:24 issue:3 pages:588-606 status: published

Intelligent road adaptive suspension system.Performance optimization with minimum actuator size.Automate design using hybrid genetic algorithm. There is an increasing demand for vehicles suitable for both on and off road driving characterized by superior comfort and handling performance. This is problematic for most suspension systems because there is a trade off balance between vibration reduction, suspension travel, actuator effort, road holding capability, as well as noise and fatigue requirements. Only in the UK every 11min a car is getting damaged because of potholes. In this paper, a method to design an intelligent suspension system with the objective to overcome the trade-off barrier using the smallest actuator is presented. An experts' based algorithm continuously monitors the road excitation in relation to the suspension travel and adapts accordingly the suspension system. It is shown that by applying genetic algorithm it is possible to optimally tune the system. However, the global optimum is hard to find due to the problem nonlinearity. A hybrid genetic algorithm that improves the probability of successfully finding the best design is presented. The simulation results show that the proposed intelligent system performs for - well known in the literature scenarios - better than others and remarkably this is achieved by reducing the actuator's size.

In this paper, a review on road friction virtual sensing approaches is provided. In particular, this work attempts to address whether the road grip potential can be estimated accurately under regular driving conditions in which the vehicle responses remain within low longitudinal and lateral excitation levels. This review covers in detail the most relevant effect-based estimation methods; these are methods in which the road friction characteristics are inferred from the tyre responses: tyre slip, tyre vibration, and tyre noise. Slip-based approaches (longitudinal dynamics, lateral dynamics, and tyre self-alignment moment) are covered in the first part of the review, while low frequency and high frequency vibration-based works are presented in the following sections. Finally, a brief summary containing the main advantages and drawbacks derived from each estimation method and the future envisaged research lines are presented in the last sections of the paper.

The latest braking system architectures for Hybrid (HEV) and Full Electric Vehicles (EV) feature the adoption of the X-by-wire solutions, namely electro-hydraulic (EHB) and electro-mechanical (EMB) braking systems, aimed at providing additional flexibility to the distinctive functions of brake blending and regeneration. Regenerative brakes still need to be supported by conventional friction brakes because of failures occurrence, fully-charged battery conditions, and unexpected variations of the tire-road friction coefficient. In order to achieve a smooth coordinated action between the regenerative and the conventional friction brakes, the brake linings coefficient of friction (BLCF) needs to be monitored. The main contribution of this work lies on the estimation of the BLCF using a tire-model-less approach. In particular, two different observer designs are proposed and compared. Whereas the proposed approach does not rely on any fixed tire modelization, the state estimation is robust against variations in the road friction characteristics and tire uncertainties caused by inflating pressure variations, wear, and aging. The functionality of the developed observers is tested in IPG CarMaker® by employing an experimentally validated EV, equipped with four onboard motors and an EHB system. Braking events are simulated at different deceleration levels on both dry and wet surfaces. Finally, the compensation function against variations in the BLCF is implemented in the EHB controller to achieve constant deceleration levels. Authors envisage that the precise knowledge of the BLCF will contribute to enhance the braking performance and to actively monitor the brake pad wear under different working conditions.

Automotive Electronics Systems Innovation Network Conference. Research & Development Session. Drift-based ADAS. Off-Road lateral dynamics enhancement. Multi-actuated Agile Electric Vehicles.

Recursive least square (RLS) with a single forgetting factor has been commonly used for parameter and state estimation of dynamical systems. In many applications such as robotics, electric vehicles, renewable energy systems, and smart-grid, accurate battery state of charge (SOC) and state of health (SOH) estimation is essential for the safe and efficient operation. To this end, the challenge lies in identifying and parameterization the temporal behavior of Lithium-Ion batteries, because their response is nonlinear and time-varying. This paper proposes a new RLS algorithm with optimum multiple adaptive forgetting factors (MAFFs) for SOC and SOH estimation of Li-ion batteries. Particle swarm intelligence is employed for identifying the system parameters. The performance of the optimum MAFF-RLS algorithm is compared to RLS with multiple fixed forgetting factors (MFFFs). Performance evaluation is carried out using the Urban Dynamometer Driving Schedule (UDDS). The simulation results indicate the better performance of MAFF-RLS algorithm compared to MFFF-RLS algorithm in terms of mean square error of SOC and internal resistance.

This paper presents a modular observer structure to estimate the tire-road forces robustly, avoiding the use of any particular tire model, and using standard signals available in current passenger vehicles. The observer consists of a feedforward longitudinal force estimation block and a hybrid lateral force estimation module formed by an Extended Kalman Filter and a Static Neural Network Structure. Road grade and bank angle are estimated using sensor fusion, where a Fuzzy Logic controller combines the outputs from a Euler Kinematic model and a Recursive Least Squares block. The proposed observer is tested and verified using the simulation software IPG CarMaker® under realistic driving situations. Lastly, the feasibility of the longitudinal force block is proved with real-time experiments.

Recent research suggests that collision mitigation on low grip surfaces might require autonomous vehicles to execute maneuvers such as drift, trail braking or Scandinavian flick. In order to achieve this it is necessary to perceive the vehicle states and their interaction with the environment, and use this information to determine the chassis limits. A first look at the virtual automotive sensing problem is provided, followed by a description of Rally driving modeling approaches. Finally, a co-pilot collision mitigation system for loose surfaces based on a Rally driver Model is proposed.