Natural and safe Human-to-Robot (H2R) object handover is a critical capability for effective Human–Robot Collaboration (HRC). However, learning a robust handover policy for this task is often hindered by the prohibitive cost of collecting physical robot demonstrations and the limitations of simplistic state representations that inadequately capture the complex dynamics of the interaction. To address these challenges, a two-stage learning framework is proposed that synthesizes substantially augmented, synthetically diverse handover demonstrations without requiring a physical robot and subsequently learns a handover policy from a rich 4D spatiotemporal flow. First, an offline, physical robot-free data-generation pipeline is introduced that produces augmented and diverse handover demonstrations, thereby eliminating the need for costly physical data collection. Second, a novel 4D spatiotemporal flow is defined as a comprehensive representation consisting of a skeletal kinematic flow that captures high-level motion dynamics and a geometric motion flow that characterizes fine-grained surface interactions. Finally, a diffusion-based policy conditioned on this spatiotemporal representation is developed to generate coherent and anticipatory robot actions. Extensive experiments demonstrate that the proposed method significantly outperforms state-of-the-art baselines in task success, efficiency, and motion quality, thereby paving the way for safer and more intuitive collaborative robots.

Rapid advancements in digitalization and artificial intelligence (AI) have catalyzed the adoption of digital twin technologies in the construction sector, enabling real-time synchronization between virtual models and physical systems. Simultaneously, on-site robotic automation has shown promise for reducing physical workloads, enhancing productivity, and contributing to sustainability goals that are key values of Industry 5.0. However, current digital twin implementations rarely incorporate multirobot construction systems, often relying on single-robot approaches or purely offline simulations. This gap hinders the realization of truly integrated construction environments that combine sensing, data analytics, wireless communications, and multirobot coordination. In response, this article proposes ConstrucTwin, a digital twin-driven multirobot construction framework designed to support complex construction tasks in real-world settings. By combining a 5G communication estimation-involved architecture and a cross-level planning strategy, ConstrucTwin streamlines interactions between physical robots and their digital counterparts. Essential tasks such as motion and task-level planning, as well as remote human-in-the-loop (HIL) oversight, are orchestrated within a single unified architecture. Through case studies involving rebar cage and brick wall construction, we demonstrate how an integrated approach to vision-based servoing and multirobot coordination enhances execution speed, precision, and scalability. The results underscore the system’s potential to advance human-centric, resilient, and sustainable construction, thereby aligning with the broader vision of Industry 5.0.

Delamination shape holds crucial information for evaluating structural safety, including its area, center, and perimeter; thus, shape prognostics has recently gained significant attention using either numerical simulations or data-driven models. Numerical approaches can capture the general trend of delamination growth while failing to account for the uncertainties arising from experimental or in-field fatigue damage growth processes. Both simulations and experiments show delamination growth along the same primary direction, but experimental observations exhibit a high degree of stochasticity in their growth rates and shape evolution that simulations cannot capture. Data-driven methods are capable of describing the actual fatigue behavior, while requiring a substantial experimental database for training. To bridge the gap between numerical simulations and complex experimental realities, we propose a framework that integrates delamination growth simulations with a data-driven approach to predict the evolution of fatigue delamination shapes. It first utilizes numerical data to train a neural ordinary differential equation (ODE)-based model that learns the gradient of the shape evolution. Subsequently, a progressive transfer learning strategy is then employed to incrementally refine the learned model using experimental observations during fatigue loading, overcoming the inherent limitations of conventional data fusion methods and enabling robust prognostics. The effectiveness of the proposed approach is demonstrated using experimental composite fatigue tests with ultrasonic C-scan monitoring, showing consistent improvements in prognostic accuracy compared with simulation-only, experiment-only, and mixed training strategies1.

Human-robot Collaboration (HRC) in unstructured environments increasingly underscores enabling robots to perceive, understand, and predict human motions to enhance its safety and efficiency, aligning with the principles of Industry 5.0. However, existing studies about human motion prediction primarily rely on deterministic methods, which struggle to capture and quantify the uncertainty in human movements, posing significant safety risks in collaborative environments. To this end, we propose a novel Fine-grained Motion Latent Diffusion (FineMLD) model tailored to generate precise and diverse human motion predictions, providing a crucial foundation for safe HRC. First, we introduce a scale-aware temporal motion latent space to capture spatio-temporal dependencies by using a time-aware human motion encoder and hierarchical motion disentanglement. The hierarchical motion disentanglement achieves a compact latent representation of the motion while enabling more precise motion decoding. Second, we perform mask-based fine-grained motion latent diffusion that incorporates trajectory-guided denoised prediction and motion mask-driven resampling modulation to refine motion predictions. The trajectory-guided denoised prediction integrates observed robot motion trajectories to guide human motion predictions, while the motion mask-driven resampling modulation aligns observed and predicted motions by using a motion mask, thereby improving the consistency and coherence of motion sequences. Finally, we validate the performance of FineMLD on a public dataset, demonstrating its state-of-the-art performance in prediction accuracy and diversity. Furthermore, we conduct physical experiments on a robotic platform to confirm its real-time applicability and robustness in a real-world laboratory setting, highlighting its potential for deployment in unstructured HRC tasks.

The manufacturing industry is undergoing a profound transformation toward smart, digital, and flexible production systems under the Industry 4.0 framework. Within this paradigm, Digital Twin (DT) serves as a key enabler, bridging physical and digital domains to simulate, analyse, and optimise manufacturing operations. Concurrently, robotic systems, enhanced by smart sensor perception, Industrial Internet of Things connectivity, and adaptive control mechanisms, are increasingly deployed to handle complex and dynamic tasks. However, the evolving demands of the modern manufacturing industry require a high degree of flexibility and responsiveness, necessitating more intelligent solutions. The Robot Digital Twin (RDT) has emerged as a transformative approach, facilitating dynamic adaptation and continuous operational improvement. This review offers a comprehensive examination of the literature on RDT in manufacturing from both technology and application perspectives, aiming to provide insight for researchers and practitioners in Industry 4.0. The paper introduces a four-layer RDT system architecture and summarises how Industry 4.0 technologies, e.g., the Industrial Internet of Things, Cloud/Edge Computing, 5 G, Virtual Reality, Modelling and Simulation, and Artificial Intelligence, converge and influence the RDT system based on this architecture. Furthermore, the review covers domain-specific and system-level applications, such as assembly, machining, grasping, material handling, human-robot interaction, predictive maintenance, and additive manufacturing systems, with an analysis of their development status. Finally, the trends, practical challenges, and future research directions for RDT systems in manufacturing are summarised at different levels.

Particle filter (PF) has proven effective for nonlinear identification scenarios; however, its performance in high-dimensional problems is often limited by the curse of dimensionality. To overcome this challenge, block particle filter (BPF) is proposed to reformulate a high-dimensional model into several blocks, so the identification of one high-dimensional system can be simplified into that for many lower-dimensional blocks. However, due to the coupling between blocks, the likelihood function for each state subgroup depends not only on its own state components (components of interest) but also on the components of its neighboring subgroups (nuisance components)—a dependency that BPF does not address. In order to extend BPF to coupled systems, we design likelihood functions, including plug-in, profile, and marginal likelihoods, that can incorporate nuisance components within each block. We demonstrate the state and parameter estimation performance of BPF with each likelihood through a numerical example of a forty-story Bouc-Wen frame structure under ground motion. We also design the BPF in a differentiable manner, integrate it into a deep learning architecture, and evaluate its performance on three datasets: the open-source Electricity Transformer Temperature (ETT) dataset, an open-source cutter wear dataset, and the AstraZeneca bearing dataset. The code is available at https://github.com/TianZLi/Likelihood-in-BPF.