This paper presents a novel GPU-enabled single- and multi-robot path planning method that integrates global evolutionary dynamic programming (EDP) with local particle swarm optimization (PSO). First, the path planning problem is formulated as a Markov decision process (MDP) using a visibility graph. Then, EDP is used to generate multiple high-quality initial paths with salient state values by solving the MDP. Lastly, the parallelized multi-population PSO is applied to optimize all initial paths near the robot simultaneously, thereby enhancing computational speed and path quality. The proposed algorithm is implemented on an edge computing device (Jetson AGX Orin) onboard a mobile robot (TurtleBot 3 Waffle Pi) for experimental verification. Additionally, an obstacle detection system is developed to recognize physical obstacles, and an indoor positioning system is used to estimate the robot’s position and orientation. The total runtime of both systems is approximately 0.06 s per frame. The optimal path is generated continuously at approximately 14 Hz, thereby facilitating successful obstacle avoidance and robot navigation in dynamic environments and demonstrating the real-time performance of obstacle and robot localization and path planning methods. Compared with other benchmarks, the proposed method significantly improves path planning robustness and quality, while also achieving substantial computational speed.

Reinforcement learning has shown strong potential for robotic path planning in dynamic environments. However, existing methods often struggle to capture complex spatiotemporal dependencies involving moving obstacles. In this work, we propose a Hybrid Transformer Path Planner (HyTra-Planner), a Parameterized Deep Q-Network (P-DQN)-based framework designed for efficient and reliable navigation in environments with both static and dynamic obstacles. HyTra-Planner presents four key innovations. First, a History Memory Generation Network (HMGN) dynamically updates temporal memory by combining current spatial features with historical information, offering a richer representation of environmental dynamics. Second, a Hybrid Transformer consists of spatial and temporal attention modules to capture both spatial and temporal dynamic patterns of obstacles. Third, an Adaptive Step Strategy allows the agent to adjust its step size according to local risks, improving navigation efficiency compared to fixed-step methods. Finally, a novel Dense Reward Mechanism is proposed to consider goal proximity, obstacle avoidance, and action efficiency altogether, which delivers continuous and informative feedback that accelerates convergence. Extensive experiments in both dynamic and hybrid environments demonstrate that HyTra-Planner achieves higher success rates, faster convergence, and more efficient path planning than state-of-the-art DQN baselines, validating its potential for real-time robotic navigation.

Unsupervised learning models have recently advanced anomaly detection, especially for complex vision-based datasets, but most still yield coarse anomaly masks with vague shapes and locations. To address these limitations, this paper presents U nsupervised S egmentation and A nomaly G radient I nterpretation ( USAGI ), a novel transformer-based unsupervised learning approach designed for accurate anomaly detection and segmentation. We propose two novel components, the Memory Transformer and Retrieval Transformer: the former builds a memory bank from normal features during training, while the latter retrieves and compares features during testing to enable fine-grained anomaly reconstruction and segmentation. USAGI achieves salient performance on the MVTec AD dataset with an AUROC of 98.2 % and on the VisA dataset (Visual Anomaly Dataset) with an AUROC of 99.5 %, 98.8 % on Real-IAD, and 96.4 % on MANTA, demonstrating its superior performance in anomaly detection and segmentation.

In autonomous driving, the ability to detect pedestrians accurately is crucial for safety. Some detectors, however, often struggle with occlusions, where pedestrians partially hidden behind objects appear incomplete and are harder to be identified accurately. To alleviate this issue, we introduce Clear3D, which make previous partially unseen pedestrians visible by effectively removing occluders to enhance 3D pedestrian detection. The first essential step in Clear3D is to accurately identify occluded regions. This is achieved by calculating the intensity variation along the LiDAR ray direction, where regions with small changes are classified as occluded. After identifying the occluded areas, they are transformed as masks and fed into the inpainting module, reconstructing the originally unseen parts. Above process, encompassing both the perception and reconstruction of occluded regions, is termed occluder removal. Finally, by fusing the recovered occlusion-free image features with LiDAR features, Clear3D generates more complete representation, enhancing the performance of downstream 3D classification and localization tasks. Extensive evaluations show that Clear3D achieves state-of-the-art accuracy across various challenging settings on KITTI, particularly in heavy occlusion environments.

Lamb waves are widely used for defect detection in structural health monitoring, and various methods are developed for Lamb wave data analysis. This paper presents an unsupervised Adversarial Transformer model for anomalous Lamb wave pattern detection by analyzing the spatiotemporal images generated by a hybrid PZT-scanning laser Doppler vibrometer (SLDV). The model includes the global attention and the local attention mechanisms, and both are trained adversarially. Given the different natures between the normal and anomalous wave patterns, global attention allows accurate reconstruction of normal wave data but is less capable of reproducing anomalous data and, hence, can be used for anomalous wave pattern detection. Local attention, however, serves as a sparring partner in the proposed adversarial training process to boost the quality of global attention. In addition, a new segment replacement strategy is also proposed to make global attention consistently extract textural contents found in normal data, which, however, are noticeably different from anomalies, leading to superior model performance. Our Adversarial Transformer model is also compared with several benchmark models and demonstrates an overall accuracy of 97.1 % for anomalous wave pattern detection. It is also confirmed that global attention and local attention in adversarial training are responsible for the superior performance of our model over the benchmark models (including the native Transformer model).

Abstract This paper presents a Spatial-Temporal Graph Convolutional Network-based Pedestrians’ behaviors Anomaly Detection system (STGCN-PAD) for grade crossings. The behaviors of pedestrians are represented in a structured manner by skeleton trajectories that are generated using a pose estimation model. The ST-GCN components are sequentially applied to capture the spatial dependencies between skeleton key points within a single video frame and the temporal relationships for each of them. Based on these features, the system reconstructs input trajectories with a constant sliding window size, and the reconstruction error is used to distinguish abnormal behaviors from those normal. To accelerate the processing of extracted multi-dimensional feature maps, an MLP-Mixer model-based reconstruction network is developed as an alternative to the traditional convolution neural network. Only trajectories of normal walking behavior are included for model training. Anomalies, such as lingering and squatting activities, can be identified as outliers by observing the magnitude of reconstruction errors. The case studies demonstrate the salient feasibility and efficiency of the proposed system, which achieves at least comparable performance (approximately 88% in the AUC evaluation metric) with several state-of-the-art approaches while using the MLP-Mixer model accelerates model inference by 10× relative to our previous effort (Song et al. in Appl Intell 53:21676–21691, 2023).

In the United States, highway-railroad grade crossings are easily congested, which not only causes significant traffic delays to travelers but also brings potential threats to the first responders for emergencies. Unfortunately, very limited research efforts have been dedicated to developing practical systems that can assess traffic conditions at overcrowded grade crossings. The main challenge in evaluating the congestion conditions at the crossings is the different instance classes (i.e., vehicle, train, and pedestrian) that need to be accurately detected, especially when densely packaged. In this study, a novel convolutional neural network (CNN) named dense traffic detection net (DTDNet) is developed. DTDNet proposes to integrate the Transformer Attention (TA) module for better modeling of global context information and the learning-to-match detection head for optimizing object detection and localization using a likelihood probability fashion. To train and test DTDNet, a unique grade crossing traffic image dataset including congested and normal traffic during both daytime and nighttime is established. Experimental results on the dataset show that the proposed DTDNet achieves the maximum mean average precision (mAP) value, 0.832, outperforming the other state-of-the-art (SOTA) models. Field test results with low mean average error (MAE), mean relative error (MRE), and root mean squared error (RMSE) which are 2.200, 1.890, and 0.280, respectively suggest the proposed model has a satisfying and robust performance in the field application under different environments.”.

The development of lab-on-a-chip systems has moved from the demonstration of individual components to a complex assembly of components. Due to the increased complexities associated with model setup and computational time requirements, current design approaches using spatial and time-resolved multiphysics modeling, though viable for component-level characterization, become unaffordable for system-level design. To overcome these limitations, we present models for the system-level simulation of fluid flow, electric field, and analyte dispersion in microfluidic devices. Compact models are used to compute the flow (pressure driven and electroosmotic) and are based on the integral formulation of the mass, momentum, and current conservation equations. An analytical model based on the method-of-moments approach has been developed to characterize the dispersion induced by combined pressure and electrokinetic-driven flow. The methodology has been validated against detailed three-dimensional (3-D) simulations and has been used to analyze hydrostatic-pressure effects in electrophoretic separation chips. A 100-fold improvement in the computational time without significantly compromising the accuracy (error less than 10%) has been demonstrated.

This paper presents composable behavioral models and a schematic-based simulation methodology to enable top-down design of electrokinetic (EK) lab-on-a-chip (LoC). Complex EK LoCs are shown to be decomposable into a system of elements with simple geometry and specific function. Parameterized and analytical models are developed to describe the electric and biofluidic behavior within each element. Electric and biofluidic pins at element terminals support the communication between adjacent elements in a simulation schematic. An analog hardware description language implementation of the models is used to simulate LoC subsystems for micromixing and electrophoretic separation. Both direct current (dc) and transient analysis can be performed to capture the influence of system topology, element sizes, material properties, and operational parameters on LoC system performance. Accuracy (relative error generally less than 5%) and speedup$(≫hbox100times)$of the schematic-based simulation methodology are demonstrated by comparison to experimental measurements and continuum numerical simulation.