Palletizing, also known as the 3D bin packing problem, is important for optimizing space utilization and automating packing processes, especially in the logistics industry. In practice, handling mixed palletizing scenarios, where a variety of boxes of different sizes are received in real time, is considerably challenging. Existing methods for solving the mixed palletizing problem often overlook practical constraints encountered in real-world applications, such as those pertaining to stability and robustness. In this paper, we propose a practical mixed palletizing manipulator system designed for structured real-world warehouse environments. Our manipulator system has two main components: a practical mixed palletizing model based on reinforcement learning (PMP-RL), which can facilitate stable and efficient box placing, and a configuration-space motion planning network (CMPNet), which can help achieve robust and efficient collision-free robot movement. The PMP-RL model is designed to maximize the pallet volume utilization while incorporating practical reward functions that enhance stability. CMPNet is used to directly predict motion trajectories in a 3D configuration space, and it facilitates real-time motion generation by effectively imitating expert-level paths. Overall, the manipulator system, comprising an automated conveyor belt, a camera-based recognition system, the PMP-RL model, and CMPNet, provide a robust and practical framework for mixed palletizing. Experiments conducted via simulations and in real-world environments have shown that the manipulator system can handle complex palletizing tasks with high efficiency and high stability.

Lifelong multiagent path finding (LMAPF) requires continuous task assignment and collision-free path planning for large robot fleets. However, existing LMAPF methods assume synchronous unit-time actions, whereas real robots execute movements asynchronously with nonuniform durations. This mismatch can lead to execution-time collisions and makes frequent replanning challenging in large-scale systems. We study LMAPF with asynchronous actions and present a framework that ensures safety under asynchronous execution and real-time scalability through coordinated path planning and action scheduling, complemented by a simple greedy task allocation. To ensure safety, the path planner eliminates cycle conflicts to prevent structural deadlocks, while a lightweight action scheduler enforces vertex precedence during execution, resolving following conflicts without explicit temporal modeling. To achieve real-time performance, the planner performs partial replanning by reusing the residual paths of nonidle agents, which are converted into a synchronous form through the scheduler’s path resynchronization, enabling consistent and efficient replanning. Experiments in warehouse environments with up to 1000 robots demonstrate that our framework more than doubles throughput compared with state-of-the-art LMAPF methods, while maintaining safe operations under asynchronous actions.

This paper presents a linear model predictive control (MPC) framework for a two-wheeled inverted pendulum with a roll joint (TWIP-R). Traditional linear quadratic regulator (LQR) approaches effectively stabilize the sagittal motion of two-wheeled inverted pendulums (TWIP), but fail to account for lateral dynamics and constraint handling during curved trajectories. Our approach decouples the system into pitch, yaw, and roll subsystems, each controlled by a dedicated linear MPC. By predicting lateral inertial forces from the reference velocity and yaw rate, the roll controller computes the zero moment point (ZMP) deviation and incorporates it into both the cost and constraints for proactive stabilization. Simulation results on a figure-eight trajectory demonstrate precise velocity tracking, robust lateral balance, and feasibility, highlighting the potential of the proposed method for dynamic, constraint-aware robot control.

Target singulation involves a rearrangement of surrounding obstacles to create space for grasping the target. However, when objects are tightly packed in a confined workspace (i.e., a limited table boundary), it is not easy to relocate obstacles. Using non-prehensile manipulation, such as push, is suitable for creating space when objects are closely placed. However, it can be risky as collisions between objects might unintentionally push them beyond the table boundary. On the other hand, prehensile motion ensures safe relocation of objects. However, the objects that can be safely grasped are limited when the packing density is too high. Thus, it may not find a rearrangement plan for target singulation. To complement both methods, we suggest using collision-free push-stack synergy for rearrangement. Collision-free push prevents objects from moving out of the boundary while efficiently relocating objects and stack creates space in advance to safely push. Furthermore, we propose a modified algorithm of Local Obstacle-based Backward Search (LOBS), which generates a global rearrangement plan using only pick-and-place actions. To evaluate our method, we set up challenging scenarios - with a packing density of 50% and up to 70 objects. Compared to LOBS, the success rate increased significantly with no meaningful increase in planning time. Additionally, Our method outperformed other baselines as well.

Deep learning models for semantic segmentation often experience performance degradation when deployed to unseen target domains unidentified during the training phase. This is mainly due to variations in image texture (i.e. style) from different data sources. To tackle this challenge, existing domain generalized semantic segmentation (DGSS) methods attempt to remove style variations from the feature. However, these approaches struggle with the entanglement of style and content, which may lead to the unintentional removal of crucial content information, causing performance degradation. This study addresses this limitation by proposing BlindNet, a novel DGSS approach that blinds the style without external modules or datasets. The main idea behind our proposed approach is to alleviate the effect of style in the encoder whilst facilitating robust segmentation in the decoder. To achieve this, BlindNet comprises two key components: covariance alignment and semantic consistency contrastive learning. Specifically, the covariance alignment trains the encoder to uniformly recognize various styles and preserve the content information of the feature, rather than removing the style-sensitive factor. Meanwhile, semantic consistency contrastive learning enables the decoder to construct discriminative class embedding space and disentangles features that are vulnerable to misclassification. Through extensive experiments, our approach outperforms existing DGSS methods, exhibiting robustness and superior performance for semantic segmentation on unseen target domains. The code is available at https://github.com/rootOyang/BlindNet.

When multiple tablewares are closely packed on a table, rearranging obstacles to make space is necessary to grasp the target, often called target singulation. Due to the nature of handling fragile tablewares (i.e. plates, bowls), we make a few assumptions for the target singulation problem. First, tableware is grasped with a multi-fingered gripper; second, rearrangement is based on prehensile motions like pick-and-place. Under these assumptions, we aim to generate a relocation plan that guarantees global optimality. Furthermore, if any relocation plan cannot singulate the target, we aim to determine it quickly. Therefore, we propose a search method that utilizes the relationship between the object and its nearby obstacles expressed in propositional logic. We define the problem as determining logical entailment (i.e., whether the target can be singulated) and expand the search tree from the target while generating an optimal relocation plan. We demonstrate the performance of our algorithm by increasing the number of objects and validate the plan in a simulation environment.

Cross-view geo-localization (CVGL) aims to determine the locations of ground-view images using corresponding aerial views. However, the inherent differences in the viewpoints and appearances between these cross-view images complicate accurate localization. Existing CVGL approaches have focused on aligning image perspectives; however, these approaches often overlook the information distortion that occurs during the alignment process. To address this challenge, we propose a novel triple-vision transformers (TriViTs) framework that fully utilizes the spatially aligned and original images. The TriViTs consist of a Siamese network and two independent networks. The Siamese network takes the polar-transformed ground or aerial images to extract the coarse features from the aligned images. Meanwhile, the two independent networks take the original images (i.e., aerial or ground images, respectively) to compensate for the distortion owing to the alignment process. Furthermore, we introduce geo-aligned triplet learning using a polar warping network (PWN) for the independent transformers to enhance the correlation between image descriptors within cross-view images. The results of extensive experiments on benchmark datasets CVUSA and CVACT prove that our method exhibits superior performance with respect to existing methods. Furthermore, the proposed simple yet effective approach demonstrates scalability across different backbones, including CNNs.

Designing robust controllers that follow a desired trajectory against unexpected internal/external disturbances presents a major challenge in the field of autonomous driving technology. In general, robust controller designs ensuring the$\textit{H}_{\infty}$performance are a popular solution. However, the$\textit{H}_{\infty}$performance only focuses on reducing the total energy of tracking errors. Sometimes, attenuating the peak value of the tracking error can be considered of equal or more importance in terms of safe driving, although this can only be achieved through the$\textit{L}_{2}-\textit{L}_{\infty}$performance. Accordingly, we propose a new controller synthesis for path-following systems in autonomous vehicles using the extended dissipativity that can consider both the maximum and total energies of the tracking error in a unified framework. In addition, autonomous vehicles usually suffer from an inevitable steering delay, which has a more adverse effect on driving stability at higher speeds. To mitigate the effect of these steering delays, we derive a new set of delay-dependent conditions for the proposed controller using the extended reciprocally convex matrix inequality. Finally, CarSim/Simulink joint-simulations under different road environments are conducted to demonstrate the effectiveness of the proposed design technique.

This paper studies delay-dependent exponential dissipative and l2-l∞filtering problems for discrete-time switched neural networks (DSNNs) including time-delayed states. By introducing a novel discrete-time inequality, which is a discrete-time version of the continuous-time Wirtinger-type inequality, we establish new sets of linear matrix inequality (LMI) criteria such that discrete-time filtering error systems are exponentially stable with guaranteed performances in the exponential dissipative and l2-l∞senses. The design of the desired exponential dissipative and l2-l∞filters for DSNNs can be achieved by solving the proposed sets of LMI conditions. Via numerical simulation results, we show the validity of the desired discrete-time filter design approach.

This paper develops a new fuzzy dynamic outputfeedback control scheme for Takagi-Sugeno (T-S) fuzzy systems with time-varying input delay and output constraints based on (Q, S, R)-α-dissipativity. The proposed controller, called a (Q, S, R)-α-dissipative output-feedback fuzzy controller, takes into consideration the abstract energy, storage function, and supply rate for the disturbance attenuation and provides a unified framework that can incorporate existing results for H∞ and passivity controllers as special cases for T-S fuzzy systems with time-varying input delay and output constraints. A dynamic parallel distributed compensator is used to design the (Q, S, R)-α-dissipative output-feedback fuzzy controller to ensure the asymptotic stability and strict (Q, S, R)-α-dissipativity of closed-loop systems described by a T-S fuzzy model that satisfies some output constraints. By employing the reciprocally convex approach, a new set of delay-dependent conditions for the desired controller is formulated in terms of the linear matrix inequality. The effectiveness and the applicability of the proposed design techniques are validated by an example of control for active suspension systems for different road conditions.

The particle filter (PF) is a popular filtering algorithm in various localization problems represented by nonlinear state-space models. Although the PF can provide accurate localization results, it often fails in localization because of the sample impoverishment phenomenon. In this paper, we propose a novel nonlinear filtering method that combines a PF with a robust filter, called a finite impulse response (FIR) filter, in order to accomplish accurate and reliable localization. The proposed filter is called the composite particle/FIR filter (CPFF). In the CPFF framework, the PF is the main filter used in normal situations. When PF failures occur, the FIR filter is used to recover the PF from failures. To detect PF failures, a new decision-making algorithm is proposed in this paper. The proposed CPFF is applied to indoor human localization using a wireless sensor network. The CPFF is accurate and reliable under conditions in which the pure PF typically exhibits degraded accuracy or failures in localization.

The biologically inspired hierarchical model for object recognition, Hierarchical Model and X (HMAX), has attracted considerable attention in recent years. HMAX is robust (i.e. shift‐ and scale‐invariant), but its use of random‐patch‐selection makes it sensitive to rotational deformation, which heavily limits its performance in object recognition. The main reason is that numerous randomly chosen patches are often orientation selective, thereby leading to mismatch. To address this issue, the authors propose a novel patch selection method for HMAX called saliency and keypoint‐based patch selection (SKPS), which is based on a saliency (attention) mechanism and multi‐scale keypoints. In contrast to the conventional random‐patch‐selection‐based HMAX model that involves huge amounts of redundant information in feature extraction, the SKPS‐based HMAX model (S‐HMAX) extracts a very few features while offering promising distinctiveness. To show the effectiveness of S‐HMAX, the authors apply it to object categorisation and conduct experiments on the CalTech101, TU Darmstadt, ImageNet and GRAZ01 databases. The experimental results demonstrate that S‐HMAX outperforms conventional HMAX and is very comparable with existing architectures that have a similar framework.

The need for accurate, fast, and reliable indoor localization using wireless sensor networks (WSNs) has recently grown in diverse areas of industry. Accurate localization in cluttered and noisy environments is commonly provided by means of a mathematical algorithm referred to as a state estimator or filter. The particle filter (PF), which is the most commonly used filter in localization, suffers from the sample impoverishment problem under typical conditions of real-time localization based on WSNs. This paper proposes a novel hybrid particle/finite impulse response (FIR) filtering algorithm for improving reliability of PF-based localization schemes under harsh conditions causing sample impoverishment. The hybrid particle/FIR filter detects the PF failures and recovers the failed PF by resetting the PF using the output of an auxiliary FIR filter. Combining the regularized particle filter (RPF) and the extended unbiased FIR (EFIR) filter, the hybrid RP/EFIR filter is constructed in this paper. Through simulations, the hybrid RP/EFIR filter demonstrates its improved reliability and ability to recover the RPF from failures.

This paper presents a Mechanically Adjustable Stiffness Actuator(MASA) for knee rehabilitation of stroke patients. The MASA is designed for safer and more effective physical human-robot interaction with patients in rehabilitation. The MASA consists of cantilever springs, a double-tripod parallel mechanism, and a torque limiter. Using the double-tripod parallel mechanism and two identical actuators, the effective length and the resting position of the cantilever springs are controlled independently. Changes of the effective length of the cantilever springs result in variation of the stiffness of the MASA. One end of each cantilever springs is attached to an axis via the torque limiter. When an external torque beyond the preset threshold is applied from and to the axis, the torque limiter is released so the axis rotates freely regardless of the position of the actuators. The MASA is used for a knee rehabilitation robot. Due to the springs and the torque limiter, physical safety of the patients is guaranteed in case of unexpected involuntary muscle activities (i.e. spasticity) during a therapy session. With changing stiffness of the MASA, the amount of assistance by the robot is possible to be adjusted.