Y. Lawrence Yao

ABSTRACT The sensitivity‐selectivity trade‐off remains a fundamental challenge in chemiresistive carbon monoxide (CO) sensing, as enhanced surface redox activity often promotes non‐specific reactions with chemically similar interferents. Single‐atom noble‐metal sensitizers improve adsorption selectivity but suffer from ultra‐low loading, limiting sensing sensitivity. Here, we report a single‐atom Pt‐activated lattice oxygen strategy that simultaneously enhances sensitivity and selectivity by engaging lattice oxygen in CO sensing. By incorporating single‐atom Pt into the SnO 2 lattice, the sensor achieves a ∼15‐fold sensitivity enhancement relative to SnO 2 , excellent selectivity against common interfering gases, and a sensing response even under oxygen‐deficient conditions. Spectroscopic analyses combined with first‐principles calculations reveal a dual functional role of lattice‐anchored Pt: single‐atom Pt establishes preferential CO adsorption sites via electronic interaction with CO, while its strong coupling with the Sn─O framework upshifts the O 2p band center, lowers the activation barrier for lattice‐oxygen‐mediated CO oxidation, and thereby enhances sensing sensitivity. Integration of the sensor into a portable CO breath analyzer further demonstrates over 8‐fold higher sensitivity than a commercial sensor under simulated exhalation conditions. This work establishes single‐atom Pt‐activated lattice oxygen strategy as an effective approach to overcome the sensitivity‐selectivity trade‐off in CO sensing.

We propose and experimentally validate an automatic urban water leakage detection architecture that leverages dark fiber links already deployed in telecommunication networks in underground conduits in the vicinity of water pipelines. The sensing stage relies on a differential-phase coherent optical time-domain reflectometry interrogator enhanced with optical pulse compression to improve sensitivity. Building on this vibration acquisition stage, automatic leakage detection algorithms are implemented by searching for leak-induced activity in the frequency domain, which is well suited to revealing leakage-related features. After acquiring a baseline calibration to characterize normal-condition vibrations at each sensing position, leakage candidates are identified by comparing distribution-based metrics computed over multiple measurements against the corresponding baseline statistics. Two automatic leakage detection strategies are developed. First, low-complexity feature-based metrics are implemented, enabling continuous monitoring with minimal computational requirements. Second, an autoencoder-based anomaly detection technique is introduced, which also relies on location-specific normal-condition calibration but reduces the dependence on prior knowledge of the expected leakage vibration signatures. A real-world field trial on an urban network demonstrates reliable detection and localization using controlled leak events generated in the field, with measurements performed over a 17 km sensing fiber and an effective spatial resolution of 2.6 m. Benchmarking against a commercial punctual electro-acoustic leak detector yields consistent trends. Overall, the proposed system could complement existing technologies by enabling automated, continuous city-scale monitoring over already deployed dark fiber infrastructure.

This work explores the relationship between the thickness of the shell of polydopamine (PDA) nanocapsules, their protein adsorption, and subsequent cellular uptake. Increasing the polymerization time of dopamine from 3 to 144 h (PCF3-PCF144) on a fructose-curcumin (CCM) template increased the capsule size from 106 to 134 nm, as measured by electron microscopy. XPS analysis revealed slight changes in the surface composition following prolonged dopamine deposition. Analysis of the protein corona using fluorescent techniques and liquid chromatography-tandem mass spectrometry (LC-MS/MS) showed that the PDA nanocapsule obtained at the shortest polymerization time (PCF3) adsorbed the most proteins and had the greatest variety, with globin and albumin being abundant. PCF3 also exhibited the highest cellular uptake in three cell lines─the breast cancer cell line MCF-7, bovine aorta endothelial cells, and the macrophage cell line RAW 264.7─while thicker shells resulted in decreased uptake. This work emphasizes that the protein corona can confer colloidal stability in serum and enhance cellular uptake.

ABSTRACT Functional polystyrene (PS) microspheres with unique size and high monodispersity play an important role in biomedical and organic synthesis. However, developing a facile strategy to prepare PS microspheres with predictable size and excellent stability remains challenging. In this study, a facile strategy is developed to prepare PS microspheres, in which sufficient sulfate radicals could be produced to initiate polymerization even without inert gas purging. 0.03 to 3 μm PS microspheres with coefficients of variation (CV) values less than 5% are prepared using optimized dispersion polymerization and emulsion polymerization. The effects of reaction conditions on the size and morphology of microspheres are studied systematically. Emulsion polymerization is utilized to prepare microsphere size less than 600 nm, and then the microspheres with size of 30 to 300 nm are fabricated by varying the copolymerization monomer concentration. Additionally, the reaction is scaled up 400 times (yield: 85%) in a 20 L reactor to further verify the large‐scale production ability of PS microspheres (CV ≤ 2%). To broaden the functionality of microspheres, microspheres with variable fluorescence intensity are prepared. The results not only provide a facile strategy for the production of narrow size distribution and high‐yield PS microspheres but also expand industrialization and downstream applications.

The complex Cu(vim) 2 (DCA) 2 catalyzes H 2 O 2 to generate the reactive oxygen species ·OH, ·O 2 − and 1 O 2 , which vigorously react with [BMIM]DCA IL to achieve an ignition delay time as low as 21 ms.

This study explores the universal hypothesis by analyzing game localization. Specifically, we compare the linguistic complexity of three popular video games: Black Myth: Wukong (China), Sekiro: Shadows Die Twice (Japan), and Red Dead Redemption 2 (USA). Using entropy and mean dependency distance as measures of lexical and syntactic complexity, we investigate whether localized versions (L2) exhibit simplification when compared with original (L1) game texts. The results show that the localized games, Black Myth and Sekiro , demonstrate greater lexical complexity and comparable or greater syntactic complexity relative to Red Dead Redemption 2 . These findings challenge the traditional simplification hypothesis in translation studies by suggesting that localization processes can lead to complexification rather than simplification. The study concludes that the multifaceted demands of localization, including cultural adaptation, technical considerations, and interactive engagement, result in unique linguistic profiles that defy expectations of universal simplification. This highlights the need for more nuanced frameworks to understand localization phenomena in the context of modern digital media.

BACKGROUND: Human herpesvirus 7 (HHV-7), a β-herpesvirus with a known tropism for CD4⁺ T cells, has not been associated with ocular surface disease. To date, no definitive cases of corneal epithelial infection directly attributed to HHV-7 have been reported. This case highlights a unique corneal manifestation following ocular surgery and provides molecular evidence supporting HHV-7 as a primary pathogen. CASE PRESENTATION: A 32-year-old man with keratoconus underwent deep anterior lamellar keratoplasty (DALK) in the left eye. Twelve days postoperatively, he developed a large geographic corneal epithelial ulcer with terminal bulbs which is classically seen in herpes simplex virus (HSV) keratitis, accompanied by conjunctival injection and ocular discomfort. Tear samples revealed HHV-7 DNA (1.1 × 10⁴ copies/sample) and U79 mRNA (1.3 × 10² copies) by quantitative polymerase chain reaction (PCR), with all other human herpesviruses testing negative. Treatment with oral valacyclovir (1000 mg TID) and topical 1% ganciclovir led to complete re-epithelialization within two weeks and a corresponding reduction in HHV-7 DNA and mRNA levels. Interestingly, multiple non-staining, grayish-white epithelial elevations resembling Thygeson’s superficial punctate keratitis emerged during recovery. HHV-7 was no longer detectable at 5 weeks, and the cornea remained clear without recurrence over 8 months. CONCLUSIONS: This case provides the first molecular and clinical evidence implicating HHV-7 as a causative agent of corneal epithelial infection. The clinical presentation closely resembled herpes simplex keratitis, and the favorable response to anti-herpetic therapy further supports its pathogenic role. HHV-7 should be considered in the differential diagnosis of dendritic or geographic keratitis even with typical terminal bulbs. Further research is warranted to clarify the corneal tropism, cellular entry mechanisms, and pathogenic potential of HHV-7.

Silver nanoplates hold significant promise for advanced electronic materials, especially in low-temperature conductive silver pastes crucial for next-generation solar cells. However, their widespread practical application, like many nanomaterials, is currently limited by insufficient production capacity and inconsistent quality inherent in conventional batch synthesis methods. To overcome these critical challenges, we developed a novel ultrasound-assisted continuous-flow synthesis method for the scalable and high-yield production of silver nanoplates. This innovative approach effectively addresses common issues such as nanoparticle deposition and pipeline clogging by leveraging ultrasonic cavitation for enhanced mixing and stable flow. Through systematic optimization of synthetic parameters-including temperature, flow rate, and seed concentration-our continuous-flow reactor achieved mass production of pure silver nanoplates at a rate of 3.8 g/h. This scaled-up system is capable of producing hundreds of grams per day. The as-prepared nanoplates demonstrate excellent electrical performance, highlighting the method's potential for industrial-scale manufacturing and significantly advancing the development of high-efficiency electronic devices.

BACKGROUND The blood-brain barrier (BBB) is essential for central nervous system (CNS) homeostasis, yet neuroinflammatory mechanisms driving BBB disruption remain poorly understood. PURPOSE To explore the oxygen-glucose deprivation/reoxygenation (OGD/R)-induced BBB dysfunction and evaluate the therapeutic effects of Guanxinning injection (GXNI), a Danshen-Chuanxiong herbal combination, targeting neuroinflammatory pathways. METHODS A 3D-BBB organoid composed of human brain microvascular endothelial cells, human astrocytes, and primary human brain microvascular pericytes was constructed, and conditions for OGD/R that simulate ischemic stroke were established. Structure and function of the in vitro BBB were evaluated by morphology, paracellular permeability, and tight junction proteins ZO-1, claudin-5, and occludin expression. In vivo, infarct volume and BBB leakage were measured in a mid-cerebral artery occlusion-induced cerebral ischemia-reperfusion injury model. RNA-seq and network pharmacology analysis were used to identify key genes and pathways for ischemic BBB disruption. HPLC-MS was performed to identify and quantify active components. Molecular docking, SPR, and molecular dynamics were performed to predict and confirm the interaction of active compounds and target proteins. RESULTS A Danshen-Chuanxiong double herbal medicine, GXNI, mitigated these effects, restoring transport capacity, reducing oxidative stress (ROS), and enhancing basement membrane components (laminin, collagen IV). In vivo, GXNI alleviated cerebral ischemia-reperfusion injury (CIRI), decreasing BBB leakage, infarct volume, and neurological deficits. The pivotal role of TLR4/NF-κB/MMP9 neuroinflammatory axis for GXNI BBB protection was identified through transcriptomic analysis and validated via immunofluorescence in BBB spheroids. Molecular docking revealed Danshen-derived salvianolic acid B (SAB) as a high-affinity MMP9 binder, confirmed by quantitative binding assays. The SAB and Chuanxiong-derived senkyunolide I (SI) combination achieved more prominent upregulation of tight junction proteins and suppression of MMP9. CONCLUSION Our findings further confirm neuroinflammation as a central driver of ischemic BBB damage and demonstrate that GXNI preserves BBB integrity by targeting TLR4/NF-κB/MMP9 signaling in 3D models and CIRI mice, with SAB-SI synergistically contributing to enhanced therapeutic efficacy.

Abstract Aircraft have received much attention because of their capability to adapt to various flight environments and complex missions. The nose cone is one of the key elements in optimising the aerodynamic shape of aircraft. A morphing nose cone (MNC) driven by a biomimetic 4-3R1U&3R sparallel mechanism is proposed in this study. Based on screw theory, the parallel mechanism’s configuration is determined, and the structure’s full-cycle degrees of freedom are concurrently confirmed. Examples in the paper demonstrate the viability of the structure by configuration synthesis, and diagrams also show the chains. This MNC is modelled after the structural design of the cicada’s abdomen and can be extended, contracted and bent. It can actively adjust its shape in response to change in the flight environments, thereby aerodynamic performance and enhancing the aircraft’s multi-mission capabilities. A scaled-down prototype is created to verify the deformation capacity of the MNC meeting the engineering requirements. Results show that the extension ratio is 36.7%, and the bending angle is 21.7°, which is better than expected. The relative error value is within a reasonable range and the extension process is incredibly stable. This research proposes new perspectives for the design of MNCs.

3D Gaussian Splatting (3D GS) achieves impressive results in novel view synthesis for small, single-object scenes through Gaussian ellipsoid initialization and adaptive density control. However, when applied to large-scale remote sensing scenes, 3D GS faces challenges: the point clouds generated by Structure-from-Motion (SfM) are often sparse, and the inherent smoothing behavior of 3D GS leads to over-reconstruction in high-frequency regions, where have detailed textures and color variations. This results in the generation of large, opaque Gaussian ellipsoids that cause gradient artifacts. Moreover, the simultaneous optimization of both geometry and texture may lead to densification of Gaussian ellipsoids at incorrect geometric locations, resulting in artifacts in other views. To address these issues, we propose PSRGS, a progressive optimization scheme based on spectral residual maps. Specifically, we create a spectral residual significance map to separate low-frequency and high-frequency regions. In the low-frequency region, we apply depth-aware and depth-smooth losses to initialize the scene geometry with low threshold. For the high-frequency region, we use gradient features with higher threshold to split and clone ellipsoids, refining the scene. The sampling rate is determined by feature responses and gradient loss. Finally, we introduce a pre-trained network that jointly computes perceptual loss from multiple views, ensuring accurate restoration of high-frequency details in both Gaussian ellipsoids geometry and color. We conduct experiments on multiple datasets to assess the effectiveness of our method, which demonstrates competitive rendering quality, especially in recovering texture details in high-frequency regions.

Building a lifelong robot that can effectively leverage prior knowledge for continuous skill acquisition remains significantly challenging. Despite the success of experience replay and parameter-efficient methods in alleviating catastrophic forgetting problem, naively applying these methods causes a failure to leverage the shared primitives between skills. To tackle these issues, we propose Primitive Prompt Learning (PPL), to achieve lifelong robot manipulation via reusable and extensible primitives. Within our two stage learning scheme, we first learn a set of primitive prompts to represent shared primitives through multi-skills pre-training stage, where motion-aware prompts are learned to capture semantic and motion shared primitives across different skills. Secondly, when acquiring new skills in lifelong span, new prompts are concatenated and optimized with frozen pretrained prompts, boosting the learning via knowledge transfer from old skills to new ones. For evaluation, we construct a large-scale skill dataset and conduct extensive experiments in both simulation and real-world tasks, demonstrating PPL’s superior performance over state-of-the-art methods.

Humans practice slow thinking before performing actual actions when handling complex tasks in the physical world. This thinking paradigm, recently, has achieved remarkable advancement in boosting Large Language Models (LLMs) to solve complex tasks in digital domains. However, the potential of slow thinking remains largely unexplored for robotic foundation models interacting with the physical world. In this work, we propose Hume: a dual-system Vision-Language-Action (VLA) model with value-guided System-2 thinking and cascaded action denoising, exploring human-like thinking capabilities of Vision-Language-Action models for dexterous robot control. System 2 of Hume implements value-Guided thinking by extending a Vision-Language-Action Model backbone with a novel value-query head to estimate the state-action value of predicted actions. The value-guided thinking is conducted by repeat sampling multiple action candidates and selecting one according to state-action value. System 1 of Hume is a lightweight reactive visuomotor policy that takes System 2 selected action and performs cascaded action denoising for dexterous robot control. At deployment time, System 2 performs value-guided thinking at a low frequency while System 1 asynchronously receives the System 2 selected action candidate and predicts fluid actions in real time. We show that Hume outperforms the existing state-of-the-art Vision-Language-Action models across multiple simulation benchmark and real-robot deployments.

Previous event-related potential (ERP) studies have demonstrated the neural specificity of cognitive processing mechanisms in scientific metaphors. This property makes semantic retrieval and extraction more difficult compared to conventional metaphors. However, the role of context in modulating the comprehension of scientific metaphors remains unclear, and there has been no analysis or categorization of abstract and difficult scientific metaphors. In this study, we used the sentence-final word paradigm to investigate the effects of different contextual conditions on the comprehension of two types of scientific metaphors. We aimed to observe (Experiment 1) whether there are any differences between the processing of the two types of scientific metaphors in the context-free condition and (Experiment 2) whether the context affects the comprehension of the two types of scientific metaphors in the contextualized condition. Additionally, we explored the modulating effects of relevant and irrelevant contexts on the two types of scientific metaphors. Both N400 and late negative component (LN) effects were found in the two experiments. The N400 analysis showed that SMF (SMF refers to scientific metaphors whose source domain and target domain have similarities in functions in present study.) evoked more negative N400 than SMS (SMS refers to scientific metaphors whose source domain and target domain have similarities in shapes in present study) in the context-free condition. The result suggests that the processing of SMF might be more difficult than that of SMS. However, in the relevant-context condition, there was no significant difference in the N400 amplitudes of the two types of scientific metaphors. In contrast, in the irrelevant-context condition, SMS elicited significantly more negative N400 than SMF. Analysis of the LN revealed no significant differences between SMS and SMF in the two experiments. The results indicate that the context might affect information extraction and retrieval, but not the late reasoning stage about scientific knowledge. Moreover, the relevant context might facilitate the comprehension of both types of scientific metaphors, whereas the irrelevant context might hinder the processing of them. More importantly, the interference seems greater for SMS.

Building a lifelong robot that can effectively leverage prior knowledge for continuous skill acquisition remains significantly challenging. Despite the success of experience replay and parameter-efficient methods in alleviating catastrophic forgetting problem, naively applying these methods causes a failure to leverage the shared primitives between skills. To tackle these issues, we propose Primitive Prompt Learning (PPL), to achieve lifelong robot manipulation via reusable and extensible primitives. Within our two stage learning scheme, we first learn a set of primitive prompts to represent shared primitives through multi-skills pre-training stage, where motion-aware prompts are learned to capture semantic and motion shared primitives across different skills. Secondly, when acquiring new skills in lifelong span, new prompts are appended and optimized with frozen pretrained prompts, boosting the learning via knowledge transfer from old skills to new ones. For evaluation, we construct a large-scale skill dataset and conduct extensive experiments in both simulation and real-world tasks, demonstrating PPL's superior performance over state-of-the-art methods.

3D Gaussian Splatting (3D GS) achieves impressive results in novel view synthesis for small, single-object scenes through Gaussian ellipsoid initialization and adaptive density control. However, when applied to large-scale remote sensing scenes, 3D GS faces challenges: the point clouds generated by Structure-from-Motion (SfM) are often sparse, and the inherent smoothing behavior of 3D GS leads to over-reconstruction in high-frequency regions, where have detailed textures and color variations. This results in the generation of large, opaque Gaussian ellipsoids that cause gradient artifacts. Moreover, the simultaneous optimization of both geometry and texture may lead to densification of Gaussian ellipsoids at incorrect geometric locations, resulting in artifacts in other views. To address these issues, we propose PSRGS, a progressive optimization scheme based on spectral residual maps. Specifically, we create a spectral residual significance map to separate low-frequency and high-frequency regions. In the low-frequency region, we apply depth-aware and depth-smooth losses to initialize the scene geometry with low threshold. For the high-frequency region, we use gradient features with higher threshold to split and clone ellipsoids, refining the scene. The sampling rate is determined by feature responses and gradient loss. Finally, we introduce a pre-trained network that jointly computes perceptual loss from multiple views, ensuring accurate restoration of high-frequency details in both Gaussian ellipsoids geometry and color. We conduct experiments on multiple datasets to assess the effectiveness of our method, which demonstrates competitive rendering quality, especially in recovering texture details in high-frequency regions.

In this paper, we claim that spatial understanding is the keypoint in robot manipulation, and propose SpatialVLA to explore effective spatial representations for the robot foundation model. Specifically, we introduce Ego3D Position Encoding to inject 3D information into the input observations of the visual-language-action model, and propose Adaptive Action Grids to represent spatial robot movement actions with adaptive discretized action grids, facilitating learning generalizable and transferrable spatial action knowledge for cross-robot control. SpatialVLA is first pre-trained on top of a vision-language model with 1.1 Million real-world robot episodes, to learn a generalist manipulation policy across multiple robot environments and tasks. After pre-training, SpatialVLA is directly applied to perform numerous tasks in a zero-shot manner. The superior results in both simulation and real-world robots demonstrate its advantage of inferring complex robot motion trajectories and its strong in-domain multi-task generalization ability. We further show the proposed Adaptive Action Grids offer a new and effective way to fine-tune the pre-trained SpatialVLA model for new simulation and real-world setups, where the pre-learned action grids are re-discretized to capture robot-specific spatial action movements of new setups. The superior results from extensive evaluations demonstrate the exceptional in-distribution generalization and out-of-distribution adaptation capability, highlighting the crucial benefit of the proposed spatial-aware representations for generalist robot policy learning. All the details and codes will be open-sourced.

Abstract Use of laser powder bed fusion (LPBF) stainless steel in corrosive environments is attractive due to material's high corrosion resistance and fine feature resolution, which is advantageous for fluidic applications. For this implementation to be optimized, LPBF stainless steel parts must have reduced susceptibility to stress corrosion cracking (SCC), a failure mode that is of high risk for stainless steels. Laser shock peening (LSP) surface processing has been used to improve SCC resistance in wrought metals and has also been used to improve other material properties of additively manufactured metals. However, LSP has yet to be investigated for the improvement of SCC behavior in LPBF stainless steel. This article demonstrates that not only does LSP improve time to crack initiation of LPBF 316L stainless steel in SCC testing but also improves SCC behavior differently when applied to different surfaces of the build. To explain these results, residual stress, texture, dislocation distribution, hardness, microstructure, and fracture surfaces are investigated, linking different hydrogen embrittlement mechanisms to each of the two build orientations as well as the peened and un-peened conditions. These results are supported by matching the observed crack morphologies to those simulated with dynamic crack modeling, thereby demonstrating the impact of residual stress and plastic versus brittle failure upon the observed outcome.

Abstract Laser shock peening (LSP) is investigated for its use in altering the electrochemical and wetting behavior of 316L stainless steel made with laser powder bed fusion (LPBF). The corrosion performance of LPBF stainless steel varies between studies and build parameters, thus motivating the search for postprocessing methods that enable wetted surface applications. Compressive surface stress has been demonstrated to reduce corrosion rate in additively manufactured metal, and LSP is known to impart compressive residual stress into metal targets. Wettability also affects corrosion behavior, and LSP induces hydrophobicity. LSP is, therefore, a promising tool for improving corrosion behavior of LPBF stainless steel. This paper examines the electrochemical properties of LPBF stainless steel before and after LSP with electrochemical impedance spectroscopy and potentiokinetic measurements. Contact angle, surface free energy, and surface finish are studied with dynamic contact angle measurements and profilometry. X-ray diffraction and energy-dispersive X-ray spectroscopy measure residual stress and surface chemistry. The top surface perpendicular to the build direction (XY) and the wall surface parallel with the build direction (XZ) are studied for all measurements due to the large differences in roughness and mechanical properties between these surfaces. LSP increases pitting potential for both XY and XZ surfaces and causes an increase to the surface electrochemical impedance. LSP also increases the contact angle of liquids on both surfaces. These changes to electrochemistry and wettability are attributed in part to surface morphology and surface chemistry alterations as well as the inducement of compressive residual stress.

Different track failures are proven to have different influences on train-track interactions, but their influence on pantograph-catenary interactions has not been investigated thus far. In this work, the influences of typical track failures on pantograph-catenary interaction dynamics are first investigated. A reduced train-track-pantograph-catenary interaction model is developed to study these effects. In this model, the long catenary is reduced to a small moving area around the moving pantograph, and this area is considered and modelled as the reduced catenary model based on the arbitrary Lagrangian–Eulerian method. The reduced catenary model and moving pantograph are formulated as the reduced pantograph-catenary interaction model, and it is further unidirectionally connected with the existing reduced train-track interaction model to formulate the reduced train-track-pantograph-catenary interaction model. The present reduced pantograph-catenary model is first validated by the EN50318-2018 standard, and the influences of track irregularity, track buckling, fastening failure, and track settlement on pantograph-catenary interaction dynamics are studied. The investigation results first show that the track settlement with a long wavelength has a significant influence on the pantograph-catenary interaction dynamics and can even cause contact loss with a good train-track interaction behaviour. Thus, it should be further mentioned in the maintenance of the track structure.

A catenary model of China high speed railway with 4-span overlap is established in this article.Through numerical simulation, the dynamic interaction of pantograph passing through the overlap at the speed of 300 km/h is calculated and analyzed.In addition, the contact force of the train running at the same speed is measured and compared with the simulation results.The simulation results are close to the line measurement results.The dynamic interaction between pantograph and catenary is significantly enhanced at overlap span, which becomes more obvious with the increase of speed.

Avoiding injury and ensuring safe evacuations of children in disasters has always been a central issue requiring close attention in policymaking. However, there is little behavioural data on children's evacuation on stairs. In this study, evacuation drills were conducted in a three-storey kindergarten in Dalian, China. The article explores the well-trained children's vertical evacuation behaviour on stairs and horizontal evacuation behaviour in other areas such as corridors and lobbies. According to the vertical behavioural evacuation data collected in this study, the mean speed of children aged 4-6 is 0.55 ± 0.12 m/s, and children of different age groups exhibit distinctive evacuation behaviours on stairs. The mean speed of children on a horizontal plane is 0.87 ± 0.22 m/s. Then, the relationship between behavioural data (movement time, density, velocity, and flow rate) is compared and analyzed. It is found that the density and flow rate of the crowd at the stairwell entrance and the exit are higher than that of adults, and the children's walking speed is relatively slow. This study can provide reference for safety design, evacuation strategies and evacuation simulation settings of multi-storey kindergartens.