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

This article presents a reactive planning and control framework to enhance the robustness of humanoid robots locomotion against external disturbances. The framework comprises two main modules, reactive planning and motion optimization. In the reactive planning module, a reactive footstep compensation strategy based on the essential motion of the linear inverted pendulum model (LIPM) is proposed. This strategy leverages the periodic motion characteristics of the LIPM, deriving the correct footstep compensation based on the conditions for model stability restoration. The module generates the zero moment point planning trajectories based on the footstep compensation. In the motion optimization module, motion optimization based on reactive planning is performed. To make motion constraint based on capture point applicable to motion optimization, the impact of different truncation points on stability constraints to determine the appropriate truncation point is quantified. The effectiveness of the proposed framework is demonstrated through experiments conducted on the humanoid robot UBTECH Walker2.

Retinal surgery requires enclosed spatial constraints to improve the safety and success of the surgery. Herein, a shared control method is proposed for master–slave robot systems, utilizing tubular guidance constraints based on a novel potential field function to optimize the commands of the surgeon. Within the tube, attractive constraints intensify with increasing task error and approach infinity at the boundary of the tube. This ensures that the surgery is confined within a closed tubular space. Haptic feedback provides force cues to inform the surgeon about the feasibility of the input commands. Theoretical derivations demonstrate that the entire closed-loop system is passive. Two simulation experiments are conducted on the ophthalmic surgery robot platform to evaluate the functionality of the proposed method. The experimental results indicate that translational errors are kept less than certain predefined values. Furthermore, the proposed method outperforms the comparison method in terms of task accuracy and efficiency.

In the rapidly advancing field of medical microrobotics, designing robots capable of addressing various challenges—such as imaging, biodegradation, and multifunctionality—is crucial. Departing from conventional research that often focuses on isolated aspects of microrobot functionality, this study presents an innovative approach to comprehensive microrobot design. Soft capsule microrobots that integrate capabilities such as magnetic navigation, autonomous maneuverability, in situ biodegradation, biosafe imaging, and drug delivery are reported. These microrobots are fabricated within the range of 20–120 μm, with a notable throughput of ≈10²–10³ microrobots per second. Furthermore, their locomotion performance has been demonstrated to remain stable for a period exceeding 10 h, all while employing real-time optical closed-loop control. The incorporation of ultrasound contrast agents not only amplifies imaging resolution but also ensures imaging contrast stability in a biological environment for over a period of 3 h. Second, the intentional integration of enzyme-loaded nanometric polymersomes establishes a self-contained, biodegradable system, accentuating the microrobots’ capacity to degrade without the addition of high enzyme concentrations. This integrated approach lays the groundwork for minimally invasive treatments toward personalized and targeted medicine.

With the advancement of artificial intelligence and internet of things, logic-in-memory (LiM) technology has garnered attention. This article presents research on LiM utilizing ferroelectric fin field-effect transistor (FinFET). Herein, the LiM characteristics of FinFET with hafnia-based switchable ferroelectric gate stack applied to the sub-3 nm future technology node are analyzed. This analysis is extended to the system level and its characteristics are observed. A compact model of the ferroelectric capacitor using Verilog-A is developed and the operation of LiM circuits such as 1-bit full adder, ternary content-addressable memory, and flip-flop by combining FinFET characteristics based on atomistic simulation with fabricated silicon-doped hafnium oxide characteristics is analyzed. Furthermore, by applying these ferroelectric devices, a power consumption reduction of 85.2% in the convolutional neural network accelerator at the system level is observed.

Artificial intelligence (AI) is increasingly used to solve multi-objective problems and reduce the turnaround times of semiconductor processes. However, only brief AI explanations are available for process/device/circuit engineers to provide holistic feedback on the manufactured results. Herein, linear/nonlinear variation decomposition (LVD/NLVD) of neural networks is demonstrated to quantitatively evaluate the influence of unit processes on the figure of merit (FoM) and co-analyze the unit process influences with device characteristic behaviors. The NLVD can evaluate the output variation from each input of neural networks in an individual sample, although neural networks are not available in an analytic form. The NLVD is successfully verified by confirming that a) the output and summation of all decomposed output variations perfectly coincide and b) the process influences on the FoM are decomposed to 6.01–54.86% more accurately compared with those of LVD in 1Y nm node dynamic random-access memory test vehicles with a baseline and split tests introducing high-k metal gates with a minimum gate length of 1 A nm node for further node scaling. The approaches identify defective processes and defect mechanisms in each sample and wafer, which enhance causal analyses for individual cases in diverse fields based on regression artificial neural networks.

Traditional analysis of highly distorted micro X-ray diffraction (μ-XRD) patterns from hydrothermal fluid environments is a time-consuming process, often requiring substantial data preprocessing and labeled experimental data. Herein, the potential of deep learning with a multitask learning (MTL) architecture to overcome these limitations is demonstrated. MTL models are trained to identify phase information in μ-XRD patterns, minimizing the need for labeled experimental data and masking preprocessing steps. Notably, MTL models show superior accuracy compared to binary classification convolutional neural networks. Additionally, introducing a tailored cross-entropy loss function improves MTL model performance. Most significantly, MTL models tuned to analyze raw and unmasked XRD patterns achieve close performance to models analyzing preprocessed data, with minimal accuracy differences. This work indicates that advanced deep learning architectures like MTL can automate arduous data handling tasks, streamline the analysis of distorted XRD patterns, and reduce the reliance on labor-intensive experimental datasets.

A biological muscle's force is nonlinearly constrained by its current state (force, length, and speed) and state history. To investigate if artificial muscles can mimic (i.e. biomimetic) the complete mechanical state spectrum of biological muscles, this study uses a novel method to characterize twisted coiled polymer actuators (TCPAs) mechanically. Thus, comprehensive and reproducible test procedures are established to verify artificial muscle biomimetics regarding stress, strain, and strain rate combinations intrinsic to biological muscle. A rheometer performs novel high-precision mechanical characterization methods to comprehensively verify biomimetic performance. Sample twist level, torque, length, force, and temperature are controlled and measured during twist-induced coiling, heatsetting/annealing, and mechanical testing. TCPAs are formed from linear low-density polyethylene monofilament. Linear low-density polyethylene (LLDPE) TCPAs generate larger stresses than biological muscle through the entire spectrum of strains—contracting more than 40%, exerting more than 0.3 MPa at rest length, and withstanding tension of 8 MPa without damage. Thus, the LLDPE TCPAs attain biological muscle performance statically, but additional tests are required to assess this dynamically. The mechanical performance of LLDPE TCPAs enables biomimetic actuation with an intelligent control and measurement system. Their high-throughput textile manufacturability positions them for advanced biomechatronic applications—including prosthetics and exoskeletons.

Automated live embryo imaging has transformed in vitro fertilization (IVF) into a data-intensive field. Unlike clinicians who rank embryos from the same IVF cycle cohort based on the embryos visual quality and determine how many embryos to transfer based on clinical factors, machine learning solutions usually combine these steps by optimizing for implantation prediction and using the same model for ranking the embryos within a cohort. Herein, it is established that this strategy can lead to suboptimal selection of embryos. It is revealed that despite enhancing implantation prediction, inclusion of clinical properties hampers ranking. Moreover, it is found that ambiguous labels of failed implantations, due to either low-quality embryos or poor clinical factors, confound both the optimal ranking and even implantation prediction. To overcome these limitations, conceptual and practical steps are proposed to enhance machine learning-driven IVF solutions. These consist of separating the optimizing of implantation from ranking by focusing on visual properties for ranking and reducing label ambiguity.

The growing interest in the flexibility and operational capabilities of soft manipulators in confined spaces emphasizes the need for precise modeling and accurate motion control. Conventional control methods encounter difficulties in modeling and involve intricate computations. This work introduces a novel deep reinforcement learning (DRL) control algorithm based on neural network modeling. Using the Whale Optimization Algorithm, an approximate dynamic model for the soft manipulator is established. The twin delayed deterministic policy gradient is employed for DRL control. Domain randomization is applied during pretraining in a simulated environment. The algorithm addresses issues related to dependency on measurement data quality and redundant mappings, outperforming other methods by 8–15 mm in control accuracy. The trained DRL controller achieves precise trajectory tracking within the soft manipulator's task space, enabling successful grasping tasks in various complex environments, including pipelines and other narrow spaces. Experimental results confirm the autonomy of our controller in performing these tasks without human intervention.

Capsule networks promise significant benefits over convolutional neural networks (CNN) by storing stronger internal representations and routing information based on the agreement between intermediate representations’ projections. Despite this, their success has been limited to small-scale classification datasets due to their computationally expensive nature. Though memory-efficient, convolutional capsules impose geometric constraints that fundamentally limit the ability of capsules to model the pose/deformation of objects. Further, they do not address the bigger memory concern of class capsules scaling up to bigger tasks such as detection or large-scale classification. Herein, a new family of capsule networks, deformable capsules (DeformCaps), is introduced to address object detection problem in computer vision. Two new algorithms associated with our DeformCaps, a novel capsule structure (SplitCaps), and a novel dynamic routing algorithm (SE-Routing), which balance computational efficiency with the need for modeling a large number of objects and classes, are proposed. This has never been achieved with capsule networks before. The proposed methods efficiently scale up to create the first-ever capsule network for object detection in the literature. The proposed architecture is a one-stage detection framework and it obtains results on microsoft common objects in context which are on par with state-of-the-art one-stage CNN-based methods, while producing fewer false-positive detection, generalizing to unusual poses/viewpoints of objects.