Excessive load sway is a critical safety concern during crane operations, exposing cranes to risks of instability and collision with surrounding objects. Existing methods for predicting load sway struggle with inefficiency and inaccuracy. Advances in robotics and automation have led to the robotisation of cranes, enhancing both safety and efficiency. This paper proposed a physics-informed neural network (PINN) for predicting the load sway of autonomous mobile cranes (AMCs) in base-moving conditions, and introduced a transfer learning (TL) framework to address complexities in AMC dynamics while reducing the need for extensive training data. Initially trained on numerically simulated data with simplified dynamics, the PINN was subsequently fine-tuned using real-world data, which included realistic dynamic uncertainties and complexities. Numerical simulations and laboratory experiments were conducted to validate the PINN’s performance. The proposed PINN accurately predicted payload motion and maintained robust performance in both numerical simulations and laboratory experiments while exhibiting superior computational efficiency, requiring only 12.5% of the time needed by traditional dynamic models for 1 s prediction windows. Furthermore, it was compared and outperformed other machine learning models, including recurrent neural networks (RNN), long short-term memory (LSTM) networks and multilayer perception (MLP). These findings indicate that the proposed PINN provides a robust and efficient solution for sensorless load sway prediction in crane operations. • Developed a PINN for autonomous mobile cranes (AMCs) load sway prediction. • Introduced a TL framework to address complexities in AMCs dynamics. • Validated through numerical simulation and laboratory experiments. • Cross-compared three normalisation methods and ten activation functions. • Benchmarked against ML methods, such as RNN, LSTM and MLP.

The necessity of human supervision and intervention during autonomous driving has long been a topic of controversial discussion. From a developer’s perspective, it is expected that users will readily adapt to well-calibrated autonomous driving systems (ADS) due to their superior performance in dynamic driving tasks (DDT) compared to conventional human-driven vehicles. However, when passengers experience an autonomous vehicle (AV), there may be an adjustment period during which they modify their behavior to accommodate the driving patterns of the ADS. Additionally, some passengers might not adapt to autonomous driving at all, highlighting potential limitations in the current ADS development strategy. This work studies the dynamics of human-automation interaction and introduces an “objective method”, which employs a Virtual Reality (VR)-enabled simulation approach for in-depth behavioral analysis concerning riders’ behavioral adaptation to autonomous driving. Specifically, we examined how participants interacted with and intervened in Level 4 ADS operating under conservative, moderate, and aggressive driving patterns in a fully autonomous environment. A realistic urban road network was recreated in VR, integrated with traffic microsimulation to generate various driving scenarios. Twenty-seven participants completed driving tasks across different AV modes, with their intervention behaviors analyzed in relation to traffic conditions and AV aggressiveness. Key findings include: (1) Participants showed higher intention to intervene but lower actual intervention rates under aggressive AV modes compared to moderate and conservative modes, suggesting quicker adaptation to more challenging driving scenarios. (2) Interventions generally proved unnecessary and sometimes detrimental to overall traffic performance in a full-AV environment. (3) Aggressive AV modes significantly improved traffic efficiency, with a 40% increase in average travel speed and a 53% reduction in waiting time. However, human interventions posed the greatest challenge to achieving optimal traffic conditions. This research provides insights into the complex dynamics of human-AV interaction and adaptation, offering valuable implications for AV interface design, implementation strategies, and public acceptance of autonomous driving technologies.

Monitoring pavement roughness is critical for minimising vehicle damages and ensuring road user safety. Conventional roughness measurement instruments are costly and limited in surveying frequency and spatial coverage. To overcome these limitations, vehicle-mounted smartphones have been adopted to measure pavement roughness based on the dynamic responses of traversing vehicles. Nonetheless, the accuracy and consistency of the current smartphone-based approaches are affected by practical factors including speed, vehicle type and mounting configuration. Existing research applied deep learning to mitigate the impact of varying practical factors, but most of them was based on simulation studies. This study introduces a method of estimating the International Roughness Index (IRI) using smartphone-collected real vehicle response. The approach leverages a multi-layer perceptron deep learning model to account for variations in practical settings. The model achieved an RMSE of 0.60 and an R2 of 0.79 when compared with the ground-truth IRI. The results showcase the deployability of the proposed data-driven method in crowdsourcing-based IRI surveying.