Trust plays a pivotal role in shaping driver interactions with autonomous vehicles (AVs), particularly in Level 3 systems that require timely human intervention during takeover requests (TORs).While prior studies have examined trust and TOR performance independently, limited work has systematically explored their intersection.This review addresses that gap by investigating how trust is formed, miscalibrated, and recovered during TOR events.Key factors such as TOR timing, modality, environmental complexity, system transparency, and individual differences are analysed in relation to both trust and takeover performance.Current trust models and measurement techniques are critically evaluated, highlighting limitations of static approaches and the emerging value of real-time trust monitoring.Guided by PRISMA 2020, a systematic literature review was conducted to screen and synthesise relevant studies.The review identifies challenges and offers design recommendations for adaptive, trust-sensitive AV systems that foster calibrated trust, ultimately improving safety, driver readiness, and overall user acceptance in automated driving contexts.

Vehicle Dynamics Models (VDMs) face a trade-off scenario between accuracy and speed. More complex models can generate more accurate predictions of vehicle state but are more computationally slow. Additionally, many VDMs rely on the explicit estimation of unknown parameters. To avoid these limitations, we propose a feed-forward Time Delay Neural Network (TDNN) which surpasses the physics-based VDMs in accuracy and speed without explicit estimation of unknown quantities. Notably, the proposed TDNN model was able to accurately predict the vehicle state on various road surfaces despite no knowledge of the tyre-road friction. The TDNN predicts the longitudinal and yaw accelerations of the vehicle with a Root Mean Square Error (RMSE) of as low as 0.0387 rad/s2.

This study reports the robust sideslip angle estimation of vehicles with an uncertain tire cornering stiffness and fading measurements. The missing measurement and possible inaccuracy in the measurement of the vehicle's yaw rate are considered by using a random variable distributed over [0, 1]. Norm-bounded uncertainties are considered in the vehicle's tire cornering stiffness. Next, the Lyapunov stability theory is used to design a sideslip angle estimator such that the filtering error dynamics is stochastically stable and the$H_{\infty}$performance criterion is met. The desired parameters of the proposed$H_{\infty}$sideslip angle estimator are gained by solving a linear matrix inequality (LMI) problem. Simulation results show that the proposed novel estimator can efficiently estimate the sideslip angle while it demonstrates robust performance to uncertainties and fading measurements.

Human Activity Recognition (HAR) is one of the essential building blocks of so many applications like security, monitoring, the internet of things and human-robot interaction. The research community has developed various methodologies to detect human activity based on various input types. However, most of the research in the field has been focused on applications other than human-in-the-centre applications. This paper focused on optimising the input signals to maximise the HAR performance from wearable sensors. A model based on Convolutional Neural Networks (CNN) has been proposed and trained on different signal combinations of three Inertial Measurement Units (IMU) that exhibit the movements of the dominant hand, leg and chest of the subject. The results demonstrate k-fold cross-validation accuracy between 99.77 and 99.98% for signals with the modality of 12 or higher. The performance of lower dimension signals, except signals containing information from both chest and ankle, was far inferior, showing between 73 and 85% accuracy.

High-G Centrifuge Systems (HCSs) are valuable tools for training aircrews and research on aviation medicine. Providing a safe and controlled environment, they are an enabler for protecting aircrews and air assets. Despite their vast applications, development of a human-rated HCS is a costly and challenging engineering project. One of the most critical steps in the development of HCSs is the experimental validation. This step has not received enough attention within the published research. This study reports evaluation and validation of an operational HCS located in the Institute for Intelligent Systems Research and Innovation (IISRI) at Deakin University, Australia. The system, owning a low-cost structure with an effective arm length of over 5m, is capable of generating a maximum sustained acceleration of 9G with an onset rate of 5G/sec. The experimental validation of system is implemented using an Advanced Wireless Human Dummy (AWHD) which is fully instrumented. The results of experimental validation show that the system can reliably generate the reference centripetal acceleration with lowest error at the human spine location.

Human walking is a complex activity with a high level of cooperation and interaction between different systems in the body. Accurate detection of the phases of the gait in real-time is crucial to control lower-limb assistive devices like exoskeletons and prostheses. There are several ways to detect the walking gait phase, ranging from cameras and depth sensors to the sensors attached to the device itself or the human body. Electromyography (EMG) is one of the input methods that has captured lots of attention due to its precision and time delay between neuromuscular activity and muscle movement. This study proposes a few Machine Learning (ML) based models on lower-limb EMG data for human walking. The proposed models are based on Gaussian Naive Bayes (NB), Decision Tree (DT), Random Forest (RF), Linear Discriminant Analysis (LDA) and Deep Convolutional Neural Networks (DCNN). The traditional ML models are trained on hand-crafted features or their reduced components using Principal Component Analysis (PCA). On the contrary, the DCNN model utilises convolutional layers to extract features from raw data. The results show up to 75% average accuracy for traditional ML models and 79% for Deep Learning (DL) model. The highest achieved accuracy in 50 trials of the training DL model is 89.5%.

Human Activity Recognition (HAR) is one of the fundamental building blocks of human assistive devices like orthoses and exoskeletons. There are different approaches to HAR depending on the application. Numerous studies have been focused on improving them by optimising input data or classification algorithms. However, most of these studies have been focused on applications like security and monitoring, smart devices, the internet of things, etc. On the other hand, HAR can help adjust and control wearable assistive devices, yet there has not been enough research facilitating its implementation. In this study, we propose several models to predict four activities from inertial sensors located in the ankle area of a lower-leg assistive device user. This choice is because they do not need to be attached to the user’s skin and can be directly implemented inside the control unit of the device. The proposed models are based on Artificial Neural Networks and could achieve up to 92.8% average classification accuracy.

Industry 5.0 is here, and human interaction experts claim that in the process of augmenting a high production/manufacturing workplace, a safety critical situation is created with the introduction of “Cobots”. A Multi-Agent Safety System (MASS) is presented as a solution in this paper which uses commercial, wearable technologies with high data sharing acceptance rates such as the Apple watch to collect and share real time ECG signals with the Cobot. Principal Component Analysis (PCA) is selected as a dimension reduction tool because it is well established and meets the requirements for reliability in the development of a human-centric, safety system. Five Machine Learning (ML) classifiers (KNN, NB, RF, DT and GBM) are used with binary classification to predict whether the human is Distracted (Event 1) or Not Distracted (Event 0) to determine if this will pose a safety risk to the Human Robot Collaboration (HRC) System. Decision Tree (DT) classifier with 4 Principal Components (PCs) is evaluated at 98% Accuracy and 99%AUC and is the recommended model for future development of the MASS. A road map is also presented to ensure the longevity of MASS while signifying the inclusion of real time data which can close the demographic data gap and help to improve the privacy, efficiency and contextual reliability of the MASS model in the Industry 5.0 workplace.

Structural design optimisation can significantly contribute to the identification of the best shape and geometry of a structure that results in lighter, stronger, and more affordable to manufacture materials for both large scale manufacturing and one-off bespoke performance components. Being both modern (evolutionary) and classic optimisation methods have had extensive focus and application in this field, an evaluation study on the performance of these methods has not been reported. This study reports a systematic comparison of the modern and classic optimisation approaches for a benchmark design optimisation problem. One algorithm will be a classic gradient-based optimisation, the second being a general Genetic Algorithm (GA) type optimisation. The results of two optimisation methods will be compared through the application of Finite Element Analysis (FEA) to evaluate both the performance of each algorithm and the real word translation of their effectiveness. The outcomes reveal that, although the gradient-based method shows better statistical results, GA can result in a superior minimum FoS of 3.5 satisfying the requirements for most common structural design applications.

Human Activity Recognition (HAR) is a crucial technology for many applications such as smart homes, surveillance, human assistance and health care. This technology utilises pattern recognition and can contribute to the development of human-in-the-loop control of different systems such as orthoses and exoskeletons. The majority of reported studies use a small dataset collected from an experiment for a specific purpose. The downsides of this approach include: 1) it is hard to generalise the outcome to different people with different biomechanical characteristics and health conditions, and 2) it cannot be implemented in applications other than the original experiment. To address these deficiencies, the current study investigates using a publicly available dataset collected for pathology diagnosis purposes to train Machine Learning (ML) algorithms. A dataset containing knee motion of participants performing different exercises has been used to classify human activity. The algorithms used in this study are Gaussian Naive Bayes, Decision Tree, Random Forest, K-Nearest Neighbors Vote, Support Vector Machine and Gradient Boosting. Furthermore, two training approaches are compared to raw data (de-noised) and manually extracted features. The results show up to 0.94 performance of the Area Under the ROC Curve (AUC) metric for 11-fold cross-validation for Gradient Boosting algorithm using raw data. This outcome reflects the validity and potential use of the proposed approach for this type of dataset.

Model Predictive Control (MPC) is a high-performing solution for Autonomous Vehicle’s (AV) control. This technique can tailor balance between various aspects of vehicle dynamics such as vehicle’s speed, acceleration and jerk. This study proposes a longitudinal controller for AV using a data-driven MPC based on human driving demonstration. A novel parameterised cost function-based MPC is designed in order to provide a general solution for different driving scenarios. This parametric cost function provides a customisable approach towards longitudinal motion generation by learning a proper set of parameter values from the user’s driving style. Instead of using any classification technique for identifying driving styles, we asked human drivers to drive with different styles and use that data directly to learn the values of the parameters. The Bayesian Optimisation (BO) approach is used to learn an optimised set of parameters minimising the gap between some carefully chosen feature values of the controller and human-generated motion. The observations of simulation show that the proposed controller is capable of generating customisable longitudinal vehicle speed, acceleration, jerk, as well as headway distance between vehicles based on a specific human driving style.

Assistive wearable exoskeleton can help to restore respective muscle functions of people with disabilities and provide support to workers within strenuous activities. The design of a practical exoskeleton should be user-friendly, compact, lightweight without any obtrusive effect on the user, which justifies the execution of optimisation steps for the design of the links and configuration of the system. Most exoskeletons struggle to accommodate a slight variation in functionality from their original design. This work addresses the design and development of a light-weight semi-active elbow exoskeleton. The main objectives consist of the development of a semi-active actuation mechanism with an adaptive exoskeleton frame and harness to accommodate varying user anthropometrics. The effectiveness of the proposed exoskeleton design has been evaluated through simulation. Furthermore, the system has been implemented in order to determine the physical feasibility and practicality through rapid fabrication prototyping. The outcomes of the study show that the exoskeleton can adjust to varying anthropometrics and assist elbow flexion, reducing stress and fatigue on the human elbow joint and muscles induced by gravity during flexion.

End to end learning for autonomous navigation and driving has become a growing research trend in both industry and academia in recent years. Its promise is in treating the whole driving pipeline as the development of a deep neural network (DNN). Its Achilles' heel is access to thousands of images required for training of the DNN. This paper comprehensively investigates the applicability of the deep transfer learning for the specific task of end to end learning of autonomous navigation. Five state of the art DNNs including ResNet, AlexNet, and Densenet are applied here for extracting features from images taken by the front-facing camera of a mobile robot. Extracted features have different information values as DNNs have different architectures and learning capabilities. These features are then processed by a multilayer fully connected neural network to estimate the robot angular velocity. Obtained results for different DNNs indicate that the transfer learning-based models show a promising performance for accurately estimating the angular velocity purely using visual information. According to obtained results, AlexNet-base model outperforms others in terms of the estimation accuracy and the performance consistency.

Path tracking controller of Autonomous Vehicles (AVs) plays an important role in improving the dynamic behaviour of the vehicle. Model Predictive Control (MPC) is one the most capable controllers that can handle multiple optimisation objectives, and accommodate the physical limits of the actuators and vehicle states to ensure safety and the other desired behaviour. As a high-potential solution, learning cost function from human demonstration can be integrated into an MPC. By learning the cost function from human demonstrations, extensive parameters tuning can be avoided, and more importantly, the controllers can be adjusted to provide desired control actions which are more natural to the human. In this study, an innovative Inverse Optimal Control (IOC) algorithm is proposed to learn a suitable cost function for the control task using collected data from human demonstration. The objective is to design a controller that generates motion which matches specific features of human-generated motion. These features include lateral acceleration, lateral velocity and deviation from the center of the lane. From the results, it is observed that the designed controller is capable of learning the desired features of human driving and implementing them while generating the appropriate control actions.

Autonomous vehicles (AVs) are considered one of the most promising solutions for enhancing road safety, saving individuals' time, and reducing energy consumption. Autonomous vehicles are still in their early stage to be publicly accepted and gain a high level of trust. They need to be comprehensively and continuously evaluated and improved through road tests which are risky, costly, and time-consuming. Motion simulators are capable of contributing to these tests by providing an immersive virtual environment and high fidelity ride experiences for subjective and objective evaluations of AVs' performance. This paper provides a simulation study on the capability of a serial motion platform, known as the Universal Motion Simulator (UMS), for emulation of an AVs' path tracking capabilities. For this purpose, a versatile path tracking model of an AV is initially introduced. The computational Multi-Body System (MBS) approach is then used to implement inverse kinematics and dynamics analyses of the UMS. The UMS is also equipped with an optimal Motion Cueing Algorithm (MCA) to emulate the motion sensation of the AV performing different manoeuvres. The results show that the UMS is an efficient tool for regenerating a realistic and high-fidelity AV ride experience when (1) curvature of the trajectory is not large and the AV does not experience large turning angles when it negotiates the path, and (2) the human motion sensation (vestibular system mathematical model) is taken into consideration in developing the MCA of the motion simulator.

This paper provides an overview on the past pieces of literature on emotion prediction systems and the different machine learning algorithms used to classify emotions. We propose a system which incorporates the emotion prediction system with a custom Smart Human Machine Interface (SHMI) for vehicle drivers to improve drive safety. This is achieved based on EEG signals and basic vehicle information's obtained from an OBD (On-Board Diagnostics) data. EEG signals are classified into four emotional states: happy, sad, relaxed and angry. In this paper, we present an initial development of the Smart Human Machine Interface (SHMI) for emotion detection for vehicle applications. To evaluate the classification of the EEG signals we use Russell's circumflex model, Higuchi Fractal Dimension (HFD), PSD (Power Spectral Density) for feature extraction and Support Vector Machines (SVM) for classification.