Advanced intelligent systems (Weinheim an der Bergstrasse, Germany)最新文献

Rapid and precise product dimension measurement is essential for enabling complete enumeration inspection, ensuring product reliability, and ultimately achieving factory automation. In particular, injection molding enables rapid and cost-effective production, making it well-suited for mass production. Thus, rapid and precise measurement is essential for inspecting the quality of all injection-molded products. However, complex 3D geometry and easily deformable property of axial fan hinder rapid and accurate measurement, thereby reducing quality control efficiency. This study introduces a convolutional neural network-based vision inspection system that can enhance the productivity and quality of injection-molded products by overcoming the limitations of traditional physical measurement methods. Consequently, the proposed model shows high performance (R-squared = ≈0.9987) for predicting both edge heights. Compared to a conventional manual measurement method, the proposed model reduces the measurement time per blade by ≈99%, and the total inspection time by ≈93.61%. Moreover, by utilizing explainable artificial intelligence, key features for prediction are identified, providing insight into why the model is capable of robust and precise measurements even in the presence of noise. The developed vision-based deflection measurement system is expected to contribute significantly to the automation of quality control of axial fans to realize the future smart injection-molding plants.

Automatic navigation of microrobot swarms driven by magnetic fields has attracted considerable attention due to potential applications in biomedical fields. However, achieving minimal loss and maintaining swarm cohesion while traversing heterogeneous landscapes over long distances remains a challenge. This article introduces a control strategy based on autonomous field-of-view (FOV) planning for navigating microrobot swarms across large workspaces that span multiple FOVs. High-resolution global images of the workspace are obtained using an image stitching method that combines phase correlation and template matching. Global path planning is accomplished with the A* algorithm, followed by local path planning utilizes the optimized informed rapidly-exploring random tree star (OI-RRT*) algorithm in each FOV to ensure swarm adaptation. The strategy also incorporates an FOV planning algorithm to optimize FOV positioning, along with a displacement platform to ensure smooth transitions between FOVs. A real-time visual feedback control system monitors both channel width and swarm position. This strategy improves swarm navigation efficiency and stability, as demonstrated through experimental validation, and holds significant potential for targeted drug delivery and other biomedical applications.

Biological systems operate through precisely coordinated interactions across multiple spatiotemporal scales, from molecules to cells, tissues, and organs. Pathologies often emerge when this homeostatic multiscale organization fails due to elements across different levels pursuing misaligned objectives, creating top-down and bottom-up cascading effects throughout the biological hierarchy. This perspective article explores how understanding these organizational failures provides valuable insights, besides for investigating fundamental processes in pathophysiology and for developing diagnostic and therapeutic strategies targeting biological organization with complex systems approaches, also for designing bioinspired artificial systems across three domains: biomimetic materials, bioinspired devices, and biomorphic computing models. This plethora of paradigms and possibilities is simplified by highlighting selected pathological mechanisms as case studies of multiscale system breakdown, namely, metabolic alterations, cancer, and neurodegenerative conditions, and how these failure modes of biological cooperation, taken in isolation and looked at in a systematic manner, present localized emergent advantages that might offer inspiration for developing adaptive and self-programmable systems, thereby expanding the pool of nature-inspired approaches beyond homeostasis.

This work presents a novel approach to autonomously sort passive beads in 3D environments using an untethered levitating magnetic microrobot. The in situ untethered magnetic weighing technique is introduced, where the classification of beads is based on the magnetic force required to levitate the microrobot and the beads. Autonomous sorting is achieved through the integration of motion control, trajectory planning, and action scheduling. Experimental validation is conducted using a nine-coil electromagnetic actuation system and a microrobot of 3 mm in size. The system demonstrates an average trajectory-following precision of 0.1 mm. The proposed magnetic weighing enables the detection and classification of carried silica beads of 0.75–1.00 mm diameter by measuring the increase in effective weight with a resolution of ≈1 μN. During the sorting process, all particles with a diameter larger than 0.9 mm are classified as heavy, and all those with a diameter smaller than 0.85 mm are classified as light, demonstrating the effectiveness of the approach. Overall, the proposed system holds significant promise for applications in biomedicine and micromanufacturing, driving innovation in autonomous microrobotics.

Existing transformer-based image captioning methods face two primary limitations: first, they struggle to adequately represent visual features from multiple regions during the encoding phase, and second, the decoder fails to effectively utilize future semantic information during the inference phase. To address these challenges, an attention-enhanced image captioning model is proposed. During the encoding phase, multigranular visual features are integrated by combining cross-attention and self-attention mechanisms, fully utilizing both grid and regional features. Additionally, a novel dense global self-attention module is introduced to enhance model performance with minimal computational cost by fully leveraging the contextual information and fine-grained details of the image. This model is particularly well-suited for biomimetic wearable devices, where real-time visual assistance plays a crucial role in enhancing the user experience. In the decoding phase, a bidirectional decoding structure with an adaptive masking module is designed to dynamically adjust the focus on past and future semantic information, enabling the model to combine historical and future context effectively for generating more accurate and relevant descriptions. Experimental results on the MSCOCO dataset show that the model outperforms the baseline, achieving a 2.1 percentage point improvement in the CIDEr metric. Comprehensive hardware evaluations on the wearable platform demonstrate real-time efficiency with minimal memory footprint, significantly outperforming state-of-the-art models in edge deployment scenarios.

In this work, fully patterned zinc tin oxide (ZTO) memristors are introduced using inkjet printing. By targeting a scalable, solution-based fabrication approach, highly stable devices with excellent reproducibility and minimal variability are achieved, using ZTO as the active layer, silver (Ag) as the top electrode, and molybdenum as the bottom electrode. The use of sustainable materials like ZTO enhances scalability and environmental compatibility, paving the way for next-generation, low-power neuromorphic computing. The devices successfully fulfill the fundamental criteria for in materia implementation of physical reservoir computing (PRC), including nonlinearity and fading memory property. The devices are successfully trained for classification tasks with MNIST handwritten dataset, achieving 89.4% accuracy and 86.5% by processing 4-bit and 5-bit input temporal sequences. The integration of printed memristors into hardware-based PRC architecture simplifies training complexity, making them particularly advantageous for energy-efficient, wearable AI systems.

Existing continuum robots with nonstretchable skeletons are monofunctional, providing only bending degrees of freedom (DOF). Additionally, their nonembedded solid structures limit application potential. In this study, inspired by origami structures, a cable-driven continuum robot featuring dual bending-contraction DOF is developed. Then, kinematic analysis to predict deformation at different cable contraction and joint parameters is conducted. Based on the kinematic characteristics of the robot, a continuum robot with five joint modules is fabricated to demonstrate its contraction and bending capabilities under no-load and loaded states. This continuum robot can reach a maximum contraction of 33.87% and a bending angle of 95.71°. By increasing the number of joints in the continuum robot, grippers with different numbers of fingers can be made for grasping objects of various shapes. Additionally, based on the embedded characteristics of the robot's cable-supported structure, a wrist rehabilitation orthosis with two joint modules is designed. This orthosis supports personalized customization, and has two rehabilitation movement modes, that is, flexion-extension (76.45° curvature) and radial-ulnar deviation (53.66° curvature). The performance and application experiments demonstrate the robot's structural conformality and potential for application in different interaction scenarios, and provide the practical guidance for cable-driven continuum robotic applications.

In order to control the posture of soft robots in dynamic environments, the movement of soft actuators such as McKibben artificial muscles must be monitored in real time. Researchers have developed various McKibben artificial muscles with embedded sensing components to achieve such monitoring. However, incorporating sensors into McKibben artificial muscles makes the system configuration more complicated and reduces the actuator's contraction rate. Furthermore, complex processing and circuitry are required, which limits applications. This study introduces a thin McKibben artificial muscle with an embedded resistive stretchable textile sensor (TES-MAM), which does not require specialized measuring instruments. TES-MAM measures displacement without additional sensing components outside the thin McKibben artificial muscle by incorporating a stretchable textile sensor inside the McKibben artificial muscle. The textile sensor exhibits excellent stretchability and does not degrade the contraction ratio. By fully embedding the textile sensor within the McKibben artificial muscle, which has an outer diameter of 4.5 mm, the structure of McKibben artificial muscle remains compact. TES-MAM provides a clear and stable resistance response during repeated pressurization and release cycles under various displacement inputs. Using TES-MAM, a single-degree-of-freedom robotic arm is constructed and its ability to measure joint posture in real time is successfully demonstrated.

A soft wearable modular assistive glove for hand assistance based on miniature foldable pouch motor unit (MFPMU) actuated by positive and negative pressure is proposed. The pouch motor consists of multiple rectangular pouches connected in series, with their centers interlinked and one side edge partially heat-sealed. It can achieve bending while simultaneously elongating, which adapts well to the extension of the skin at finger joints when rotating. Based on the principle of virtual work, the relationship between the output torque and the bending angle is established and experimentally verified. Test results indicate that the MFPMU can attain a maximum output torque of 240 mN m. The bending angle of the bending segment can achieve 62° under 10 kPa. The fingertip force is 2.34 N under 50 kPa for a single finger in the original state. The proposed glove utilizes a modular design, allowing users to replace damaged units within 30 s. The proposed soft wearable modular assistive glove can assist the hand to perform different hand motions and grasp objects with various shapes, dimensions, and weights. Finally, during grasping a wooden block, the soft wearable modular assistive glove demonstrates an average increase of 56% in each finger's fingertip force.

Scalable data collection from challenging locations, such as forests or bridges, is essential for biodiversity and environmental monitoring, as well as infrastructure and industrial inspection. Robots can collect this data, by placing sensors with uncrewed aerial vehicles (UAVs), perching UAVs, or climbing robots. All require adhesion to substrates with varying roughnesses, from tree bark to concrete and glass. Unfortunately, common adhesion methods are specialized for specific substrates, don't generalize to different surfaces, or leave behind harmful residue. This work presents the novel use of gelatin-based hydrogels as biodegradable and water-soluble adhesives for reversible adhesion to different surfaces for robotic environmental monitoring. The hydrogel adheres through heating, attaching, and cooling. The hydrogel is released by heating again, and any residue can be washed away with water. To correctly dimension the adhesive, factors affecting the maximum pull-off force are experimentally characterized. The versatility of the adhesive is shown through adhesion to different surfaces. Only 0.1 g is shown to support at least 20 N. Finally, the adhesion method is validated on three robotic monitoring applications: sensor placement, UAV perching, and a climbing robot. These tests demonstrate the utility of biodegradable gelatin hydrogels as adhesives for robotic monitoring applications in natural and industrial settings.