Place recognition is the foundation for autonomous systems to achieve independent decision-making and secure operation. It is also crucial in tasks such as loop closure detection and global localization in Simultaneous Localization and Mapping (SLAM) technology. Existing LiDAR-based place recognition (LPR) methods use raw point cloud representations or multifarious point cloud representations as inputs, as well as employ convolutional neural networks or transformer architectures. However, the recently proposed Mamba deep learning model combined with State Space Models (SSMs) has enormous potential in long sequence modeling. Therefore, we have developed a novel place recognition network OverlapMamba, which represents input range views (RVs) as sequences. In a novel way, we use a stochastic reconstruction method to establish shifted state space models to compress the visual representation. Extensive experiments on three public datasets demonstrate that OverlapMamba achieves competitive performance with real-time inference speed, which effectively detects loop closure even when traversing previously visited locations from different directions, indicating its strong place recognition ability and real-time efficiency.

Robust local feature detection and description are foundational tasks in computer vision. Existing methods primarily rely on single appearance cues for modeling, leading to unstable keypoints and insufficient descriptor discriminability. In this paper, we propose a multi-cue guided local feature learning framework that leverages semantic and geometric cues to synergistically enhance detection robustness and descriptor discriminability. Specifically, we construct a joint semantic-normal prediction head and a depth stability prediction head atop a lightweight backbone. The former leverages a shared 3D vector field to deeply couple semantic and normal cues, thereby resolving optimization interference from heterogeneous inconsistencies. The latter quantifies the reliability of local regions from a geometric consistency perspective, providing deterministic guidance for robust keypoint selection. Based on these predictions, we introduce the Semantic-Depth Aware Keypoint (SDAK) mechanism for feature detection. By coupling semantic reliability with depth stability, SDAK reweights keypoint responses to suppress spurious features in unreliable regions. For descriptor construction, we design a Unified Triple-Cue Fusion (UTCF) module, which employs a semantic-scheduled gating mechanism to adaptively inject multi-attribute features, improving descriptor discriminability. Extensive experiments on four benchmarks validate the effectiveness of the proposed framework. The source code and pre-trained model will be available at: https://github.com/yiyscut/GESS.git.

The prevailing paradigm of perceptive humanoid locomotion relies heavily on active depth sensors. However, this depth-centric approach fundamentally discards the rich semantic and dense appearance cues of the visual world, severing low-level control from the high-level reasoning essential for general embodied intelligence. While monocular RGB offers a ubiquitous, information-dense alternative, end-to-end reinforcement learning from raw 2D pixels suffers from extreme sample inefficiency and catastrophic sim-to-real collapse due to the inherent loss of geometric scale. To break this deadlock, we propose GeoLoco, a purely RGB-driven locomotion framework that conceptualizes monocular images as high-dimensional 3D latent representations by harnessing the powerful geometric priors of a frozen, scale-aware Visual Foundation Model (VFM). Rather than naive feature concatenation, we design a proprioceptive-query multi-head cross-attention mechanism that dynamically attends to task-critical topological features conditioned on the robot's real-time gait phase. Crucially, to prevent the policy from overfitting to superficial textures, we introduce a dual-head auxiliary learning scheme. This explicit regularization forces the high-dimensional latent space to strictly align with the physical terrain geometry, ensuring robust zero-shot sim-to-real transfer. Trained exclusively in simulation, GeoLoco achieves robust zero-shot transfer to the Unitree G1 humanoid and successfully negotiates challenging terrains.

LiDAR-based place recognition (LPR) is a key component of visual localization and autonomous driving. Although LiDAR data are usually preprocessed by motion undistortion, which can greatly reduce scene distortion caused by sensor motion, 3-dimensional (3D) point clouds in complex scenes still show inherent unstructured sparsity. Projecting 3D point clouds into 2D range view (RV) images can provide structured feature representations that are more suitable for neural networks and deep learning methods. However, this strategy has two main limitations: 1) the important geometric details are removed by the projection, impairing the effective extraction and representation of scene features. 2) unpredictable occlusions and viewpoint changes alter the observed scene structure and often lead to incorrect place matches. To address these limitations, we propose a LPR network that combines topological information extracted by persistent homology (PH) with RV features to compensate for missing structural information and produce more stable descriptors. Specifically, PH-stable features are added as a topology-aware sampling channel for the RV feature to integrate more stable and discriminative robust geometric cues. Next, we introduce a spatiotemporal fusion module that links topological and RV features across consecutive frames, thereby improving spatiotemporal consistency of the representation and thus reducing the impact of occlusions and rotations. Compared with OverlapMamba, PHMRNet can better handle spatial, temporal, and geometric features, improving the robustness of the descriptor. Experiments on public datasets show that PHMRNet outperforms existing methods in terms of accuracy and robustness under the same driving conditions and LiDAR scanning rate.

Developing an amphibious hip powered exoskeleton (AHPE) for lower limb augment in amphibious environments is crucial, which is facing the challenges, such as mechanical design, effective human-exoskeleton interaction, and simulating scenarios, for performance validations. To tackle these issues, this work proposes a design of a cable-driven AHPE named AmphiHip Exo for the hip flexion/extension assisting. To achieve an effective human-exoskeleton interaction, a physics-informed neural network named PILaN is proposed to realize hip motion intention perception (HMIP)-based control of the AmphiHip Exo. Trained by a self-collected small-scale dataset containing hip states of human activities, the proposed PILaN can automatically proceed HMIP, including joint trajectory and velocity estimations, with real-time dynamics parameter estimations and the Lagrangian dynamics model. Prototype tests are conducted on various wearers under four designed task scenarios: daily handling, climbing stairs, underwater gait assistance, and water rescue, and the performances are validated according to our designed evaluation metrics: HMIP accuracy, power assisting effect (PAE), and mechanical reliability. Experimental results prove the accuracy and efficiency of PILaN, the PAE among different wearers and task scenarios, and the waterproof effect as well as battery endurance, demonstrating the feasibility of the designed AmphiHip Exo.

Natural language provides an intuitive way to express spatial intent in geospatial applications. While existing localization methods often rely on dense point cloud maps or high-resolution imagery, OpenStreetMap (OSM) offers a compact and freely available map representation that encodes rich semantic and structural information, making it well suited for large-scale localization. However, text-to-OSM (T2O) localization remains largely unexplored. In this paper, we formulate the T2O global localization task, which aims to estimate accurate 2 degree-of-freedom (DoF) positions in urban environments from textual scene descriptions without relying on geometric observations or GNSS-based initial location. To support the proposed task, we introduce TOL, a large-scale benchmark spanning multiple continents and diverse urban environments. TOL contains approximately 121K textual queries paired with OSM map tiles and covers about 316 km of road trajectories across Boston, Karlsruhe, and Singapore. We further propose TOLoc, a coarse-to-fine localization framework that explicitly models the semantics of surrounding objects and their directional information. In the coarse stage, direction-aware features are extracted from both textual descriptions and OSM tiles to construct global descriptors, which are used to retrieve candidate locations for the query. In the fine stage, the query text and top-1 retrieved tile are jointly processed, where a dedicated alignment module fuses textual descriptor and local map features to regress the 2-DoF pose. Experimental results demonstrate that TOLoc achieves strong localization performance, outperforming the best existing method by 6.53

Trajectory prediction of surrounding agents is a prerequisite for safe planning and decision making in autonomous driving. Without high-definition (HD) maps, sensor-derived bird's-eye-view (BEV) features provide no explicit lane topology or drivable-area priors, making it inherently difficult to ground each agent in its surrounding scene context. Moreover, physical feasibility remains difficult to capture through data-driven learning alone, as kinematic constraints on agent motion cannot be explicitly encoded without structured supervision. Existing map-free predictors extract scene context in an agent-agnostic manner through a single fusion step and treat physical constraints only as output-level penalties, leaving both challenges unaddressed. We propose SIPTraj, a map-free trajectory prediction framework that jointly addresses scene grounding and physical feasibility. SIPTraj introduces a Hierarchical Agent-Scene Encoder (HASE) progressively grounding each agent in agent-guided scene evidence and refining inter-agent relations within the scene-grounded space. To tackle physical infeasibility in predicted trajectories, we develop a Physics-Guided Iterative Decoder (PGID). It conditions decoding on instantaneous kinematic states, propagating physical supervision into internal representations rather than output trajectories alone. Extensive experiments on nuScenes and Argoverse 2 Sensor show that SIPTraj surpasses prior map-free predictors and strong map-based baselines without any HD map at inference. Our code will be released as open-source.

Tactile and kinesthetic perceptions are crucial for human dexterous manipulation, enabling reliable grasping of objects via proprioceptive sensorimotor integration. For robotic hands, even though acquiring such tactile and kinesthetic feedback is feasible, establishing a direct mapping from this sensory feedback to motor actions remains challenging. In this article, we propose a novel glove-mediated tactile–kinematic perception–prediction framework for grasp skill transfer from human intuitive and natural operation to robotic execution based on imitation learning, and its effectiveness is validated through generalized grasping tasks, including those involving deformable objects. First, we integrate a data glove to capture tactile and kinesthetic data at the joint level. The glove is adaptable for both human and robotic hands, allowing data collection from natural human hand demonstrations across different scenarios. It ensures consistency in the raw data format, enabling evaluation of grasping for both human and robotic hands. Second, we establish a unified representation of multimodal inputs based on graph structures with polar coordinates. We explicitly integrate the morphological differences into the designed representation, enhancing the compatibility across different demonstrators and robotic hands. Furthermore, we introduce the tactile–kinesthetic spatio-temporal graph networks, which leverage multidimensional subgraph convolutions and attention-based long short-term memory (LSTM) layers to extract spatio-temporal features from graph inputs to predict node-based states for each hand joint. These predictions are then mapped to final commands through a force-position hybrid mapping. Comparative experiments and ablation studies demonstrate that our approach surpasses other methods in grasp success rate, finger coordination, contact force management, and both grasp and computational efficiency, achieving results most akin to human grasping.

Reconstructing large-scale RGB point clouds is an important task in robotics, supporting perception, navigation, and scene understanding. Despite advances in LiDAR inertial visual odometry (LIVO), its performance remains highly sensitive to extrinsic calibration. Meanwhile, 3D vision foundation models, such as VGGT, suffer from limited scalability in large environments and inherently lack metric scale. To overcome these limitations, we propose LiDAR-VGGT, a novel framework that couples LiDAR inertial odometry with the state-of-the-art VGGT model through a two-stage coarse-to-fine fusion pipeline: First, a pre-fusion module robustly estimates per-session VGGT poses and RGB point clouds with a coarse metric scale. Then, a post-fusion module enhances cross-modal 3D similarity transformation, using bounding-box-based regularization to reduce scale distortions caused by inconsistent FOVs between LiDAR and camera sensors. Extensive experiments across multiple datasets demonstrate that LiDAR-VGGT achieves improved visual quality and density compared to VGGT-based methods and LIVO baselines. The implementation of our method and the proposed novel RGB point cloud evaluation toolkit are released as open source.

LiDAR place recognition (LPR) plays a vital role in autonomous navigation. However, existing LPR methods struggle to maintain robustness under adverse weather conditions such as rain, snow, and fog, where weather-induced noise and point cloud degradation impair LiDAR reliability and perception accuracy. To tackle these challenges, we propose an Iterative Task-Driven Framework (ITDNet), which integrates a LiDAR Data Restoration (LDR) module and a LiDAR Place Recognition (LPR) module through an iterative learning strategy. These modules are jointly trained end-to-end, with alternating optimization to enhance performance. The core rationale of ITDNet is to leverage the LDR module to recover the corrupted point clouds while preserving structural consistency with clean data, thereby improving LPR accuracy in adverse weather. Simultaneously, the LPR task provides feature pseudo-labels to guide the LDR module's training, aligning it more effectively with the LPR task. To achieve this, we first design a task-driven LPR loss and a reconstruction loss to jointly supervise the optimization of the LDR module. Furthermore, for the LDR module, we propose a Dual-Domain Mixer (DDM) block for frequency-spatial feature fusion and a Semantic-Aware Generator (SAG) block for semantic-guided restoration. In addition, for the LPR module, we introduce a Multi-Frequency Transformer (MFT) block and a Wavelet Pyramid NetVLAD (WPN) block to aggregate multi-scale, robust global descriptors. Finally, extensive experiments on Weather-KITTI, Boreas, and our proposed Weather-Apollo datasets demonstrate that, ITDNet outperforms existing LPR methods, achieving state-of-the-art performance in adverse weather.

Tracking any point (TAP) is a fundamental yet challenging task in computer vision, requiring high precision and long-term motion reasoning. Recent attempts to combine RGB frames and event streams have shown promise, yet they typically rely on synchronous or non-adaptive fusion, leading to temporal misalignment and severe degradation when one modality fails. We introduce TAPFormer, a transformer-based framework that performs asynchronous temporal-consistent fusion of frames and events for robust and high-frequency arbitrary point tracking. Our key innovation is a Transient Asynchronous Fusion (TAF) mechanism, which explicitly models the temporal evolution between discrete frames through continuous event updates, bridging the gap between low-rate frames and high-rate events. In addition, a Cross-modal Locally Weighted Fusion (CLWF) module adaptively adjusts spatial attention according to modality reliability, yielding stable and discriminative features even under blur or low light. To evaluate our approach under realistic conditions, we construct a novel real-world frame-event TAP dataset under diverse illumination and motion conditions. Our method outperforms existing point trackers, achieving a 28.2

Forecasting how human hands move in egocentric views is critical for applications like augmented reality, human-robot policy transfer, and service/assistive technologies. Recently, several hand trajectory prediction (HTP) methods have been developed to generate future possible hand waypoints, which still suffer from insufficient prediction targets, inherent modality gaps, entangled hand-head motion, and limited validation in downstream tasks. To address these limitations, we present Uni-Hand, a universal hand motion forecasting framework considering multi-modal input, multi-dimensional and multi-target prediction patterns, and multi-task affordances for downstream applications. We harmonize multiple modalities by vision-language fusion, global context incorporation, and task-aware text embedding injection, to forecast hand waypoints in both 2D and 3D spaces. A novel dual-branch diffusion is proposed to concurrently predict human head and hand movements, capturing their motion synergy in egocentric vision. By introducing target indicators, the prediction model can forecast the specific joint waypoints of the wrist or the fingers, besides the widely studied hand center points. In addition, we enable Uni-Hand to additionally predict hand-object interaction states (contact/separation) to facilitate downstream tasks better. To incorporate comprehensive downstream task evaluations in the literature, we build novel benchmarks to assess the real-world applicability of hand motion forecasting algorithms. The experimental results on multiple publicly available datasets and our newly proposed benchmarks demonstrate that Uni-Hand achieves the state-of-the-art performance in multi-dimensional and multi-target hand motion forecasting. Extensive validation in multiple downstream tasks also presents its impressive human-robot policy transfer to enable robotic manipulation, and effective feature enhancement for action anticipation/recognition.

LiDAR-camera extrinsic calibration is essential for multi-modal data fusion in robotic perception systems. However, existing approaches typically rely on handcrafted calibration targets (e.g., checkerboards) or specific, static scene types, limiting their adaptability and deployment in real-world autonomous and robotic applications. This article presents the first self-supervised LiDAR-camera extrinsic calibration network that operates in an online fashion and eliminates the need for specific calibration targets. We first identify a significant generalization degradation problem in prior methods, caused by the conventional single-sided data augmentation strategy. To overcome this limitation, we propose a novel double-sided data augmentation technique that generates multi-perspective camera views using estimated depth maps, thereby enhancing robustness and diversity during training. Built upon this augmentation strategy, we design a dual-path, self-supervised calibration framework that reduces the dependence on high-precision ground truth labels and supports fully adaptive online calibration. Furthermore, to improve cross-modal feature association, we replace the traditional dual-branch feature extraction design with a difference map construction process that explicitly correlates LiDAR and camera features. This not only enhances calibration accuracy but also reduces model complexity. Extensive experiments conducted on five public benchmark datasets, as well as our own recorded dataset, demonstrate that the proposed method significantly outperforms existing approaches in terms of generalizability.

LiDAR-based place recognition (LPR) is essential for global localization and loop-closure detection in large-scale SLAM systems. Existing methods typically construct global descriptors from Range Images or BEV representations for matching. BEV is widely adopted due to its explicit 2D spatial layout encoding and efficient retrieval. However, conventional BEV representations rely on simple statistical aggregation, which fails to capture fine-grained geometric structures, leading to performance degradation in complex or repetitive environments. To address this, we propose MPTF-Net, a novel multi-view multi-scale pyramid Transformer fusion network. Our core contribution is a multi-channel NDT-based BEV encoding that explicitly models local geometric complexity and intensity distributions via Normal Distribution Transform, providing a noise-resilient structural prior. To effectively integrate these features, we develop a customized pyramid Transformer module that captures cross-view interactive correlations between Range Image Views (RIV) and NDT-BEV at multiple spatial scales. Extensive experiments on the nuScenes, KITTI and NCLT datasets demonstrate that MPTF-Net achieves state-of-the-art performance, specifically attaining a Recall@1 of 96.31% on the nuScenes Boston split while maintaining an inference latency of only 10.02 ms, making it highly suitable for real-time autonomous unmanned systems.

Robust environmental perception is fundamental for embodied agents performing long-term navigation and loop closure detection. While Multimodal Large Language Models (MLLMs) promise superior semantic understanding for dealing with severe environmental changes, their high latency and computational cost hinder deployment on resource-constrained robotic platforms. To bridge this gap, we present a novel zero-shot framework that leverages language-guided reasoning for efficient robotic visual place recognition (VPR). We introduce Multi-Scale Semantic-Difference Pruning (MS-SDP), a context-aware perception mechanism that dynamically filters dynamic noise (e.g., moving vehicles) and compresses visual tokens based on cross-modal semantic relevance. Furthermore, by bypassing intermediate text generation through structured Chain-of-Thought prompting, our method enables the agent to perform direct, end-to-end similarity reasoning. Extensive experiments demonstrate that our approach not only achieves state-of-the-art zero-shot performance in challenging navigation scenarios where traditional sensors fail, but also delivers a 223x speedup, making MLLM-driven perception feasible for real-time autonomous systems.

Multi-Layer Perceptron (MLP) models are the foundation of contemporary point cloud processing. However, their complex network architectures obscure the source of their strength and limit the application of these models. In this article, we develop a two-stage abstraction and refinement (ABS-REF) view for modular feature extraction in point cloud processing. This view elucidates that whereas the early models focused on ABS stages, the more recent techniques devise sophisticated REF stages to attain performance advantages. Then, we propose a High-dimensional Positional Encoding (HPE) module to explicitly utilize intrinsic positional information, extending the "positional encoding" concept from Transformer literature. HPE can be readily deployed in MLP-based architectures and is compatible with transformer-based methods. Within our ABS-REF view, we rethink local aggregation in MLP-based methods and propose replacing time-consuming local MLP operations, which are used to capture local relationships among neighbors. Instead, we use non-local MLPs for efficient non-local information updates, combined with the proposed HPE for effective local information representation. We leverage our modules to develop HPENets, a suite of MLP networks that follow the ABS-REF paradigm, incorporating a scalable HPE-based REF stage. Extensive experiments on seven public datasets across four different tasks show that HPENets deliver a strong balance between efficiency and effectiveness. Notably, HPENet surpasses PointNeXt, a strong MLP-based counterpart, by 1.1% mAcc, 4.0% mIoU, 1.8% mIoU and 0.2% Cls. mIoU, with only 50.0%, 21.5%, 23.1%, 44.4% of FLOPs on ScanObjectNN, S3DIS, ScanNet, and ShapeNetPart, respectively.

A two-stage pipeline is widely used in Visual Place Recognition (VPR) for mobile robotics to improve robustness under challenging conditions. It typically consists of two steps: descriptor generation and re-ranking. In descriptor generation, a backbone network generates descriptors for the query and database images. Then, re-ranking performs three steps: 1) top-k candidate descriptors are selected from the database; 2) these candidates are aggregated into a global descriptor; 3) a re-ordered list of candidates is returned according to similarity scores against the global descriptor. However, current re-ranking methods suffer from two critical limitations: 1) False Positive Aggregation: Under texture variations, positive and negative database samples may produce similar descriptors. As a result, some negative samples may be selected into re-ranking, contaminating the aggregation and hence incorrect top-k similarity scores. 2) Limited Similarity Correction: Most re-ordering is typically non-differentiable so that no matching error can be back-propagated to train the re-ranking stage to learn calibrated similarity scores during network training. To address these issues, we propose a novel re-ranking framework called Dynamic Graph Attention Optimal Transport (DGAOT), comprising two key re-ranking modules built upon optimal transport theory. First, a Dynamic Graph Aggregation (DGA) module is designed to convert the pairwise affinities of the top-k query–candidate graph into an OT-constrained joint coupling, down-weight the falsely selected candidates and aggregate the descriptors under consistent distributional patterns, improving the robustness to texture variations. Second, Differentiable Reordering Layer (DRL) - a differentiable score calibration layer is designed to perform differentiable score re-allocation via Sinkhorn iterations. Finally, a new Coarse-to-fine Average Precision (CFAP) loss is proposed to supervise DGA and DRL, enabling end-to-end re-ranking. Extensive experiments show that DGAOT achieves high retrieval performance, by about 5.3% relative and 3.35% absolute improvement in R@1 on average.

Open-world instance classification in LiDAR point clouds aims to discover and categorize instances of unknown classes unseen during training, without access to their ground-truth labels. The task is extremely challenging since only knownclass labels are available for training. Existing LiDAR classification methods often depend on restrictive priors, including a predefined number of novel classes or additional annotations for unseen objects, which limits their applicability in continually changing environments. In this paper, we propose LOIC, an open-world instance classification framework for LiDAR point clouds with instance-level contrastive learning. LOIC first obtains candidate unknown instances and their voxel features through open-set semantic segmentation and ellipsoidal clustering. The introduced Classification Feature Extraction Head module (CFEH) then transforms point-wise features into discriminative cluster-based instance representations. The proposed Classification with Geometry-aware High-dimensional Features module (CGHF) further integrates orientation-robust geometric cues with the learned representations. Unknown instances are continuously assigned to discovered categories or initialized as novel categories according to their distances from dynamically maintained class prototypes, without requiring the unknown-class number in advance. We establish an evaluation benchmark on the nuScenes, SemanticKITTI, and in-house Campus datasets. Experiments show that LOIC outperforms state-of-the-art approaches adapted from the visual and feature-clustering domains, achieving improvements by 3.0, 4.8, and 7.9 percentage points in PQ, mIoUu, and mPreu, respectively. The source code of our method will be publicly available at https://github.com/nubot-nudt/LOIC.

Explainability and transparent decision-making are essential for the safe deployment of autonomous driving systems. Scene captioning summarizes environmental conditions and risk factors in natural language, improving transparency, safety, and human--robot interaction. However, most existing approaches target structured urban scenarios; in off-road environments, they are vulnerable to single-modality degradations caused by rain, fog, snow, and darkness, and they lack a unified framework that jointly models structured scene captioning and path planning. To bridge this gap, we propose Wild-Drive, an efficient framework for off-road scene captioning and path planning. Wild-Drive adopts modern multimodal encoders and introduces a task-conditioned modality-routing bridge, MoRo-Former, to adaptively aggregate reliable information under degraded sensing. It then integrates an efficient large language model (LLM), together with a planning token and a gate recurrent unit (GRU) decoder, to generate structured captions and predict future trajectories. We also build the OR-C2P Benchmark, which covers structured off-road scene captioning and path planning under diverse sensor corruption conditions. Experiments on OR-C2P dataset and a self-collected dataset show that Wild-Drive outperforms prior LLM-based methods and remains more stable under degraded sensing. The code and benchmark will be publicly available at https://github.com/wangzihanggg/Wild-Drive.

High-fidelity global static maps reconstructed from 3-D LiDAR data play a crucial role in precise spatial measurement tasks such as urban spatial modeling, large-scale environmental measurement, and remote sensing. However, dynamic objects may introduce "ghost trail" during map construction, which significantly degrades map quality and limits the execution of downstream tasks. Existing dynamic removal methods often struggle to balance computational efficiency and accuracy. To address this issue, this article presents a stratified and tiered multiscale descriptor framework for LiDAR-based static environment measurement, STEP-Mover, which efficiently filters dynamic points from the global map to preserve its high-fidelity features. The proposed method generates a voxel-based multiscale descriptor and progressively localizes dynamic regions in a coarse-fine-refine manner by leveraging feature differences between the query scan and the prior map. To compensate for blind spots in the query scan, a high point retrieval strategy is introduced to retrieve static points outside the field of view (FOV). Finally, the coarsely extracted ground points are used as priors, and union-find-based connectivity clustering is applied to refine ground point recovery. Extensive comparative experiments are conducted on the SemanticKITTI, HeLiMOS, and self-collected M2UD datasets. Quantitative and qualitative results demonstrate that, compared with other baselines, the proposed method has an advantage in balancing static preservation and dynamic removal while maintaining efficiency. Our code is open-sourced at https://yaepiii.github.io/STEP-Mover/