Shenzhen Academy of Robotics

A vision-driven driver monitoring system plays a vital role of vehicular cyber-physical systems (VCPS) to guarantee the driving safety. Recent advances focus on modeling a deep learning-based method to realize the driver monitoring system, which benefits from the powerful capability of data-driven feature extraction. Although the acceptable performances of driver state monitoring methods are achieved, there is still a gap between the emerged techniques and actual application scenarios. First, the human-centric visual appearances are not involved comprehensively to represent the driver states, resulting in ignoring the contextual interaction of behaviors. Second, the inherent uncertainty of driver situation is not considered, while the unreliable samples would lead to the untrustworthy results. In this paper, we focus on a vision-based driver state monitoring method, where a trustworthy driver state perception (TDSP) is proposed via human-centric contextual interaction-driven evidential vision-language fusion in VCPS. Specifically, a vision-language model-based architecture is first modified in temporal dimension to represent the visual human-centric contextual interactions, while a vision-language consistency loss is designed to mitigate the gap between visual and textual representations. Then, an evidence-based learning method is introduced to jointly conduct the classification and uncertainty estimation for driver states. Furthermore, to model the human-centric contextual interactions towards the evidence-based paradigm comprehensively, Dempster-Shafer theory-based combination rule is introduced to fuse the visual and textual representations. Extensive experiments are conducted on two public benchmarks, where the superiority of TDSP is demonstrated compared with the state-of-the-art methods. The superior performance of TDSP to recognize dangerous states are 85.41% and 83.12% in terms of accuracy and F1 score, which outperforms the state-of-the-art methods by 4.68% and 3.99%. Moreover, we validate the reliability of TDSP against the noisy data for VCPS. The code will be public at https://github.com/w64228013/TDSP.

Synthetic jets, generated through the periodic suction and ejection of fluid without net mass addition, offer distinct benefits, such as compactness, ease of integration, and independence from external fluid sources. These characteristics make them well-suited for flow control and convective heat transfer applications. However, conventional single-actuator configurations are constrained by limited jet formation, narrow surface coverage, and diminished effectiveness in the far field. This review critically evaluates the key limitations and explores four advanced configurations developed to mitigate them: dual-cavity synthetic jets, single-actuator multi-orifice jets, coaxial synthetic jets, and synthetic jet arrays. Dual-cavity synthetic jets enhance volume flow rate and surface coverage by generating multiple vortices and enabling jet vectoring, though they remain constrained by downstream vortex diffusion. Single-actuator multi-orifice designs enhance near-field heat transfer through multiple interacting vortices, yet far-field performance remains an issue. Coaxial synthetic jets improve vortex dynamics and overall performance but face challenges at high Reynolds numbers. Synthetic jet arrays with independently controlled actuators offer the greatest potential, enabling jet vectoring and focusing to enhance entrainment, expand spanwise coverage, and improve far-field performance. By examining key limitations and technological advances, this review lays the foundation for expanded use of synthetic jets in practical engineering applications.

Electronics inspection in production and maintenance lines faces distinct challenges. Fixed-focus imaging thrives in standardized production layouts but faces challenges with unpredictable irregularities in maintenance lines, such as component layouts and geometries. To address this, researchers explored adaptive platforms, particularly soft programmable surfaces for pose control. However, most existing solutions rely on light-sensitive feedback, and lack passive adaptability and integrated backlit illumination, which limits their effectiveness. This article introduces the villi-like adaptive leveling and backlit illumination soft surface (VALBS), a soft robotic platform that integrates passive compliance, active pneumatic three degree of freedom pose control, and embedded backlit illumination, enabling robust inspection without altering camera hardware. Building on this, the soft actuated multiadaptive inspection (SAMI) platform is constructed, comprising a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">${\rm{2 \times 2}}$</tex-math></inline-formula> array of VALBS units. Each VALBS adjusts the object’s pose through pneumatic actuation, guided by an analytical control model and visual feedback. Experiments show SAMI achieves 100% contour detection accuracy and 85% –100% pose adjustment success rate on five representative layouts pattern of printed circuit board assembly along with a range of commercially available mainboards, validating the platform’s effectiveness in handling geometric and lighting condition variability in maintenance-line inspection.

With the advancement of humanoid robotics, the challenge of robotic manipulation under environmental constraints has drawn increasing attention. Conventional industrial grippers often perform poorly when grasping flat objects. To address this issue, this paper proposes a novel WYZY Hand that integrates a Watt mechanism with a double parallelogram mechanism. Designed with a three-phalanx structure mimicking the human finger, the hand supports three grasping modes: parallel pinching, scooping, and adaptive grasping, enabling composite manipulation strategies. Through passive downward contact with a desktop, the hand automatically switches from parallel pinching to scooping mode upon reaching a certain height via finger-table interaction, thereby effectively gripping flat objects. The adaptive mode accommodates desktops of various shapes, while compound grasping allows handling of irregular objects. This paper presents kinematic and dynamic analyses, establishes a detailed dynamic model for each linkage, and examines the force transmission mechanism. The range of grasping force is derived, and safety validation is conducted. Experimental results demonstrate that the proposed hand achieves effective and stable performance in grasping diverse objects, offering a viable solution to manipulation tasks under environmental constraints.

The development of intelligent robots seeks to seamlessly integrate them into the human world, providing assistance and companionship in daily life and work, with the ultimate goal of achieving human-robot symbiosis. This requires robots with intelligent interaction abilities to work naturally and effectively with humans. However, current robotic simulators fail to support real human participation, limiting their ability to provide authentic interaction experiences and gather valuable human feedback essential for enhancing robotic capabilities. In this paper, we introduce SymBridge, the first human-in-the-loop cyber-physical interactive system designed to enable the safe and efficient development, evaluation, and optimization of human-robot interaction methods. Specifically, we employ augmented reality technology to enable real humans to interact with virtual robots in physical environments, creating an authentic interactive experience. Building on this, we propose a novel robotic interaction model that generates responsive, precise robot actions in real time through continuous human behavior observation. The model incorporates multi-resolution human motion features and environmental affordances, ensuring contextually adaptive robotic responses. Additionally, SymBridge enables continuous robot learning by collecting human feedback and dynamically adapting the robotic interaction model. By leveraging a designed system architecture and modules, SymBridge builds a bridge between humans and robots, as well as between cyber and physical spaces, providing a natural and realistic interaction experience while facilitating the continuous evolution of robotic intelligence. Extensive experiments, user studies, and robot testing demonstrate the system’s promising performance and highlight its potential to significantly advance research on human-robot symbiosis.

ABSTRACT Performing complex tasks in open environments remains challenging for robots, even when using large language models (LLMs) as the core planner. Many LLM‐based planners are inefficient due to their large number of parameters and prone to inaccuracies because they operate in open‐loop systems. We think the reason is that only applying LLMs as planners is insufficient. In this study, we propose Dadu‐Embodied ( Dadu‐E) , a robust closed‐loop planning framework for embodied Artificial Intelligence (AI) robots. Specifically, Dadu‐E is equipped with a relatively lightweight LLM, a set of encapsulated robot skill instructions, a robust feedback system, and memory augmentation. Together, these components enable Dadu‐E to (i) actively perceive and adapt to dynamic environments, (ii) optimize computational costs while maintaining high performance, and (iii) recover from execution failures using its memory and feedback mechanisms. By seamlessly integrating a lightweight LLM, encapsulated robot skill instructions, closed‐loop feedback, and memory augmentation, Dadu‐E establishes a robust and extensible embodied AI robotic framework capable of stable operation in real‐world environments. Extensive experiments on real‐world and simulated tasks show that Dadu‐E achieves task success rates comparable to embodied AI robots with larger models as planners like COME‐Robot, while reducing computational requirements by . Users are encouraged to explore our system at https://rlc-lab.github.io/dadu-e/ .

Predominant fault diagnosis in industrial robots depends on dedicated vibration or acoustics sensors. However, their practical deployment is often limited by installation constraints, susceptibility to environmental noise, and cost considerations. Applying Energy-Based Maintenance (EBM) principles to achieve enhanced fault diagnosis under practical industrial conditions, we propose a hybrid deep learning framework, the Multi-head Graph Attention Network (MGAT) with Multi-scale CNNBiLSTM Fusion (MGAT-MCNNBiLSTM) for industrial robots. This approach obviates the need for additional dedicated sensors, effectively mitigating associated deployment complexities. The framework embodies four core innovations: (1) Based on the EBM paradigm, motor current is established as the most effective and practical choice for enabling cost-efficient and scalable industrial robot fault diagnosis. A corresponding dataset of motor current has been acquired from industrial robots operating under diverse fault scenarios. (2) An integrated MGAT-MCNNBiLSTM architecture that synergistically models multiscale local features and complex dynamics through its MCNNBiLSTM module while capturing nonlinear interdependencies via MGAT. This comprehensive feature representation enables robust and highly accurate fault detection. (3) The study found that the application of spectral preprocessing techniques yields a marked and statistically significant enhancement in diagnostic performance. A comprehensive and systematic analysis was undertaken to uncover the underlying reasons for this observed performance improvement. (4) To emulate challenging industrial settings and cost-sensitive implementations, noise signal injection was employed to evaluate model robustness in high-electromagnetic-interference environments and low-cost, low-resolution ADC implementations. Experimental validation on real-world industrial robot datasets demonstrates that MGAT-MCNNBiLSTM achieves a superior diagnostic accuracy of 90.7560%. This performance marks a significant absolute improvement of 1.51–8.55% over competing models, including LCNNBiLSTM, SCNNBiLSTM, MCCBiLSTM, GAT, and MGAT. Under challenging noise and low-resolution conditions, the proposed model consistently outperforms CNNBiLSTM variants, GAT, and MGAT with an improvement of 1.37–10.26% and enhanced industrial utility and deployment potential.

The integration of Artificial Intelligence (AI) into contemporary art curation is revolutionizing how exhibitions are designed, artworks are selected, and audiences interact with art. Through technologies such as machine learning, natural language processing (NLP), and computer vision, AI enables curators to analyze vast collections, predict visitor preferences, and construct dynamic exhibition experiences. This paper examines the dual nature of AI in art curation—its potential to enhance creativity and efficiency, and the ethical, technical, and philosophical challenges it introduces. Drawing upon case studies from global museums and digital art platforms, the research identifies key application domains, evaluates curatorial outcomes, and analyzes stakeholders’ perceptions of AI-assisted practices. Quantitative data from 30 institutions are supplemented by qualitative analysis to explore how AI reshapes curatorial authority, audience engagement, and the meaning-making process. Findings reveal that while AI fosters innovation and accessibility, it also risks algorithmic bias, cultural homogenization, and diminished human interpretive agency. The paper concludes that sustainable AI curation requires a human–machine collaborative framework emphasizing transparency, inclusivity, and cultural sensitivity.