arXiv subjects

Sicheng Wang

Publications and source records attributed to Sicheng Wang.

At least 19 recordsLinked to original sources

arXiv2026

SafeAtlas-VL: Beyond Binary Multimodal Safety with Large-Scale Data and Guard Models

Multimodal safety moderation requires distinguishing risks arising from visual content, user intent, and assistant behavior. Existing safeguards, however, are typically trained for a single judgment target and reduce safety assessment to a binary decision. Consequently, risk becomes difficult to compare across a multimodal interaction, and ambiguous cases are obscured. We introduce SafeAtlas-VL, a dataset of 1.5M training instances that places image-, request-, and response-level judgments on a five-level ordered scale. We curate a broad collection of safety-relevant data from both real-world and synthetic sources and apply a disagreement-aware annotation procedure. The resulting dataset spans 15 harm categories and 55 fine-grained subcategories, covering a broad range of multimodal safety scenarios. We also construct SafeAtlas-Bench, a held-out set of 5,000 instances for evaluating five-level predictions and continuous risk scores. Upon this dataset, we train the SafeAtlas Guard series of models via target-conditioned tuning for multimodal safety detection. Our models not only perform five-way classification of safety levels but also map safety to continuous scores through a soft cumulative ordinal head. Experimental results demonstrate that guard models trained on our dataset exhibit strong generalization: even without using the training sets of other benchmarks, they achieve competitive performance on the corresponding test sets. Notably, our 8B model attains the overall best performance, outperforming the previous SOTA by approximately 4% in F1 score. Code, data, and models are released to support further research. Warning: this paper contains example data that may be offensive, harmful, graphic, or disturbing.

cs.AI↗

arXiv2026

Rate-Splitting Multiple Access Enabled Probabilistic Semantic Communication in UAV Networks

This article proposes an uncrewed aerial vehicle (UAV) downlink semantic communication framework, where probabilistic knowledge graphs (PKGs) are employed to model user equipment (UE) semantics and decompose semantic information into shared and private components. Leveraging the capability of rate-splitting multiple access (RSMA) in addressing such semantic structures, a PKG-assisted RSMA transmission scheme is developed to efficiently deliver multi-user semantic information under severe energy constraints and fast-varying UAV channels. To characterize the strongly coupled energy costs of communication, computation, and flight, a weighted energy minimization problem is formulated to jointly optimize the UAV trajectory, power allocation, beamforming design, and semantic compression ratio. The resulting non-convex problem is efficiently solved using an iterative semantic-aware weighted energy optimization (SWEO) algorithm that integrates Lagrangian dual decomposition and successive convex approximation. Furthermore, a semantic accuracy metric is proposed to quantify the reliability of reconstruction by assigning importance-based weights to informative KG triples. Extensive simulation results verify that the proposed framework achieves superior energy efficiency, enhanced semantic preservation, and consistently better performance than conventional RSMA, non-orthogonal multiple access (NOMA), and space division multiple access (SDMA) schemes in benchmarks across various network parameters.

eess.SP↗

arXiv2026

Emergent Misalignment Can Be Induced by Sycophancy and Reversed via Alignment Gating

Prior work has shown that fine-tuning large language models on malicious or incorrect outputs in narrow domains can induce broad misalignment and harmful behavior, a phenomenon known as emergent misalignment. However, efficient methods for reversing such misalignment remain limited. In this work, we make two contributions. First, we identify sycophancy fine-tuning, i.e., training models to passively agree with users' incorrect opinions, as a previously underexplored driver of emergent misalignment, and show that it induces broad and severe misaligned behavior. Second, we propose Alignment Gating, an efficient method for reversing emergent misalignment that inserts learnable and controllable gates into the model during fine-tuning. Through fine-tuning, these gates learn to identify the internal representations responsible for unsafe responses. Thus, amplifying or suppressing these representations then exacerbates or mitigates EM, respectively. We further find that alignment gating module exhibits strong generalization: gating weights obtained from narrow-domain fine-tuning substantially suppress broad-domain misaligned behavior while preserving the model's general capabilities.

cs.CL↗

arXiv2026

FG-TreeSeg: Flow-Guided Tree Crown Segmentation without Instance Annotations

Individual tree crown segmentation is an important task in remote sensing for forest biomass estimation and ecological monitoring. However, accurate delineation in dense, overlapping canopies remains a bottleneck. While supervised deep learning methods suffer from high annotation costs and limited generalization, emerging foundation models (e.g., Segment Anything Model) often lack domain knowledge, leading to under-segmentation in dense clusters. To bridge this gap, we propose FG-TreeSeg, a training-free framework for tree crown instance segmentation that transfers flow-based delineation from biomedical imaging to remote sensing. By modeling tree crowns as star-convex objects within a topological flow field using Cellpose-SAM, the FG-TreeSeg framework forces the separation of touching tree crown instances based on vector convergence. Experiments on the NEON and BAMFOREST datasets and visual inspection demonstrate that our framework generalizes robustly across diverse sensor types and canopy densities, which can offer a training-free solution for tree crown instance segmentation and labels generation.

cs.CV↗

arXiv2026

Cross-Modal Urban Sensing: Evaluating Sound-Vision Alignment Across Street-Level and Aerial Imagery

Environmental soundscapes convey substantial ecological and social information regarding urban environments; however, their potential remains largely untapped in large-scale geographic analysis. In this study, we investigate the extent to which urban sounds correspond with visual scenes by comparing various visual representation strategies in capturing acoustic semantics. We employ a multimodal approach that integrates geo-referenced sound recordings with both street-level and remote sensing imagery across three major global cities: London, New York, and Tokyo. Utilizing the AST model for audio, along with CLIP and RemoteCLIP for imagery, as well as CLIPSeg and Seg-Earth OV for semantic segmentation, we extract embeddings and class-level features to evaluate cross-modal similarity. The results indicate that street view embeddings demonstrate stronger alignment with environmental sounds compared to segmentation outputs, whereas remote sensing segmentation is more effective in interpreting ecological categories through a Biophony--Geophony--Anthrophony (BGA) framework. These findings imply that embedding-based models offer superior semantic alignment, while segmentation-based methods provide interpretable links between visual structure and acoustic ecology. This work advances the burgeoning field of multimodal urban sensing by offering novel perspectives for incorporating sound into geospatial analysis.

cs.CV↗

arXiv2026

Parallel Simulation of Contact and Actuation for Soft Growing Robots

Soft growing robots, commonly referred to as vine robots, have demonstrated remarkable ability to interact safely and robustly with unstructured and dynamic environments. It is therefore natural to exploit contact with the environment for planning and design optimization tasks. Previous research has focused on planning under contact for passively deforming robots with pre-formed bends. However, adding active steering to these soft growing robots is necessary for successful navigation in more complex environments. To this end, we develop a unified modeling framework that integrates vine robot growth, bending, actuation, and obstacle contact. We extend the beam moment model to include the effects of actuation on kinematics under growth and then use these models to develop a fast parallel simulation framework. We validate our model and simulator with real robot experiments. To showcase the capabilities of our framework, we apply our model in a design optimization task to find designs for vine robots navigating through cluttered environments, identifying designs that minimize the number of required actuators by exploiting environmental contacts. We show the robustness of the designs to environmental and manufacturing uncertainties. Finally, we fabricate an optimized design and successfully deploy it in an obstacle-rich environment.

cs.RO↗

arXiv2025

IVGAE-TAMA-BO: A novel temporal dynamic variational graph model for link prediction in global food trade networks with momentum structural memory and Bayesian optimization

Global food trade plays a crucial role in ensuring food security and maintaining supply chain stability. However, its network structure evolves dynamically under the influence of geopolitical, economic, and environmental factors, making it challenging to model and predict future trade links. Effectively capturing temporal patterns in food trade networks is therefore essential for improving the accuracy and robustness of link prediction. This study introduces IVGAE-TAMA-BO, a novel dynamic graph neural network designed to model evolving trade structures and predict future links in global food trade networks. To the best of our knowledge, this is the first work to apply dynamic graph neural networks to this domain, significantly enhancing predictive performance. Building upon the original IVGAE framework, the proposed model incorporates a Trade-Aware Momentum Aggregator (TAMA) to capture the temporal evolution of trade networks, jointly modeling short-term fluctuations and long-term structural dependencies. A momentum-based structural memory mechanism further improves predictive stability and performance. In addition, Bayesian optimization is used to automatically tune key hyperparameters, enhancing generalization across diverse trade scenarios. Extensive experiments on five crop-specific datasets demonstrate that IVGAE-TAMA substantially outperforms the static IVGAE and other dynamic baselines by effectively modeling temporal dependencies, while Bayesian optimization further boosts performance in IVGAE-TAMA-BO. These results highlight the proposed framework as a robust and scalable solution for structural prediction in global trade networks, with strong potential for applications in food security monitoring and policy decision support.

cs.AI↗

arXiv2025

DynaMIC: Dynamic Multimodal In-Context Learning Enabled Embodied Robot Counterfactual Resistance Ability

The emergence of large pre-trained models based on natural language has breathed new life into robotics development. Extensive research has integrated large models with robots, utilizing the powerful semantic understanding and generation capabilities of large models to facilitate robot control through natural language instructions gradually. However, we found that robots that strictly adhere to human instructions, especially those containing misleading information, may encounter errors during task execution, potentially leading to safety hazards. This resembles the concept of counterfactuals in natural language processing (NLP), which has not yet attracted much attention in robotic research. In an effort to highlight this issue for future studies, this paper introduced directive counterfactuals (DCFs) arising from misleading human directives. We present DynaMIC, a framework for generating robot task flows to identify DCFs and relay feedback to humans proactively. This capability can help robots be sensitive to potential DCFs within a task, thus enhancing the reliability of the execution process. We conducted semantic-level experiments and ablation studies, showcasing the effectiveness of this framework.

cs.RO↗

arXiv2025

APT*: Asymptotically Optimal Motion Planning via Adaptively Prolated Elliptical R-Nearest Neighbors

Optimal path planning aims to determine a sequence of states from a start to a goal while accounting for planning objectives. Popular methods often integrate fixed batch sizes and neglect information on obstacles, which is not problem-specific. This study introduces Adaptively Prolated Trees (APT*), a novel sampling-based motion planner that extends based on Force Direction Informed Trees (FDIT*), integrating adaptive batch-sizing and elliptical $r$-nearest neighbor modules to dynamically modulate the path searching process based on environmental feedback. APT* adjusts batch sizes based on the hypervolume of the informed sets and considers vertices as electric charges that obey Coulomb's law to define virtual forces via neighbor samples, thereby refining the prolate nearest neighbor selection. These modules employ non-linear prolate methods to adaptively adjust the electric charges of vertices for force definition, thereby improving the convergence rate with lower solution costs. Comparative analyses show that APT* outperforms existing single-query sampling-based planners in dimensions from $\mathbb{R}^4$ to $\mathbb{R}^{16}$, and it was further validated through a real-world robot manipulation task. A video showcasing our experimental results is available at: https://youtu.be/gCcUr8LiEw4

cs.RO↗

arXiv2025

RoboBERT: An End-to-end Multimodal Robotic Manipulation Model

Embodied intelligence seamlessly integrates vision, language, and action.~However, most multimodal robotic models rely on massive fine-tuning, incurring high time and hardware costs.~To address this, we introduce RoboBERT, an end-to-end multimodal manipulation model built around a novel two-stage training paradigm.~In the first stage, we freeze most of the vision encoder and train with a single "standard" instruction phrasing, allowing the model to focus on stable policy learning via a CNN-based diffusion policy.~In the second stage, we unfreeze all modules and inject diverse natural language variants, rapidly aligning varied instructions to the already-learned policy without destabilizing performance.~We further employ systematic data augmentations to enhance robustness against visual perturbations.~Without relying on auxiliary datasets, RoboBERT achieves new state-of-the-art (SOTA) mean episode lengths of 4.52 on the CALVIN ABCD-D benchmark and 3.79 on the ABC-D benchmark using only language-labeled expert demonstrations and a comparatively lightweight architecture.Real-robot trials on a 6-DOF manipulator confirm higher success rates than comparable methods trained on identical data.These results demonstrate that our data-augmentation-enhanced two-stage training paradigm delivers efficient, scalable, and broadly applicable performance for multimodal robotic systems.

cs.RO↗

arXiv2025

A GAN-Enhanced Deep Learning Framework for Rooftop Detection from Historical Aerial Imagery

Precise detection of rooftops from historical aerial imagery is essential for analyzing long-term urban development and human settlement patterns. Nonetheless, black-and-white analog photographs present considerable challenges for modern object detection frameworks due to their limited spatial resolution, absence of color information, and archival degradation. To address these challenges, this research introduces a two-stage image enhancement pipeline based on Generative Adversarial Networks (GANs): image colorization utilizing DeOldify, followed by super-resolution enhancement with Real-ESRGAN. The enhanced images were subsequently employed to train and evaluate rooftop detection models, including Faster R-CNN, DETReg, and YOLOv11n. The results demonstrate that the combination of colorization with super-resolution significantly enhances detection performance, with YOLOv11n achieving a mean Average Precision (mAP) exceeding 85\%. This signifies an enhancement of approximately 40\% over the original black-and-white images and 20\% over images enhanced solely through colorization. The proposed method effectively bridges the gap between archival imagery and contemporary deep learning techniques, facilitating more reliable extraction of building footprints from historical aerial photographs. Code and resources for reproducing our results are publicly available at \href{https://github.com/Pengyu-gis/Historical-Aerial-Photos}{github.com/Pengyu-gis/Historical-Aerial-Photos}.

cs.CV↗

arXiv2025

Physics-Grounded Differentiable Simulation for Soft Growing Robots

Soft-growing robots (i.e., vine robots) are a promising class of soft robots that allow for navigation and growth in tightly confined environments. However, these robots remain challenging to model and control due to the complex interplay of the inflated structure and inextensible materials, which leads to obstacles for autonomous operation and design optimization. Although there exist simulators for these systems that have achieved qualitative and quantitative success in matching high-level behavior, they still often fail to capture realistic vine robot shapes using simplified parameter models and have difficulties in high-throughput simulation necessary for planning and parameter optimization. We propose a differentiable simulator for these systems, enabling the use of the simulator "in-the-loop" of gradient-based optimization approaches to address the issues listed above. With the more complex parameter fitting made possible by this approach, we experimentally validate and integrate a closed-form nonlinear stiffness model for thin-walled inflated tubes based on a first-principles approach to local material wrinkling. Our simulator also takes advantage of data-parallel operations by leveraging existing differentiable computation frameworks, allowing multiple simultaneous rollouts. We demonstrate the feasibility of using a physics-grounded nonlinear stiffness model within our simulator, and how it can be an effective tool in sim-to-real transfer. We provide our implementation open source.

cs.RO↗

arXiv2024

Quantifying Global Food Trade: A Net Caloric Content Approach to Food Trade Network Analysis

As the global population and the per capita demand for resource intensive diets continues to grow, the corresponding increase in food demand challenges the global food system, enhancing its reliance on trade. Most previous research typically constructed either unweighted networks or weighted solely by tonnage to represent food trade, and focused on bilateral trade relationships between pairs of countries. This study investigates the properties of global food trade constructed in terms of total food calories associated with all the main food products exchanged along each trade link (edge of the food trade network). Utilizing data from the Food and Agriculture Organization between 1986 and 2022, we construct a directed, weighted network of net caloric flows between countries. This approach highlights the importance of considering nutritional value in discussions of food security and trade policies, offering a more holistic view of global food trade dynamics. Our analysis reveals significant heterogeneity in trade patterns, with certain countries emerging as major exporters or importers of food calories. Moreover, we employ network measures, including network connectivity, network heterogeneity, network modularity, and node correlation similarity, to elucidate the structural dynamics of global net food calorie trade networks that are relevant to the stability and resilience of the global food system. Our work provides a more nuanced understanding of global food trade dynamics, emphasizing the need for comprehensive strategies to enhance the resilience and sustainability of food trade networks.

econ.GN↗

arXiv2024

Anisotropic Stiffness and Programmable Actuation for Soft Robots Enabled by an Inflated Rotational Joint

Soft robots are known for their ability to perform tasks with great adaptability, enabled by their distributed, non-uniform stiffness and actuation. Bending is the most fundamental motion for soft robot design, but creating robust, and easy-to-fabricate soft bending joint with tunable properties remains an active problem of research. In this work, we demonstrate an inflatable actuation module for soft robots with a defined bending plane enabled by forced partial wrinkling. This lowers the structural stiffness in the bending direction, with the final stiffness easily designed by the ratio of wrinkled and unwrinkled regions. We present models and experimental characterization showing the stiffness properties of the actuation module, as well as its ability to maintain the kinematic constraint over a large range of loading conditions. We demonstrate the potential for complex actuation in a soft continuum robot and for decoupling actuation force and efficiency from load capacity. The module provides a novel method for embedding intelligent actuation into soft pneumatic robots.

cs.RO↗

arXiv2024

How Does Diverse Interpretability of Textual Prompts Impact Medical Vision-Language Zero-Shot Tasks?

Recent advancements in medical vision-language pre-training (MedVLP) have significantly enhanced zero-shot medical vision tasks such as image classification by leveraging large-scale medical image-text pair pre-training. However, the performance of these tasks can be heavily influenced by the variability in textual prompts describing the categories, necessitating robustness in MedVLP models to diverse prompt styles. Yet, this sensitivity remains underexplored. In this work, we are the first to systematically assess the sensitivity of three widely-used MedVLP methods to a variety of prompts across 15 different diseases. To achieve this, we designed six unique prompt styles to mirror real clinical scenarios, which were subsequently ranked by interpretability. Our findings indicate that all MedVLP models evaluated show unstable performance across different prompt styles, suggesting a lack of robustness. Additionally, the models' performance varied with increasing prompt interpretability, revealing difficulties in comprehending complex medical concepts. This study underscores the need for further development in MedVLP methodologies to enhance their robustness to diverse zero-shot prompts.

cs.CV↗

arXiv2024

Conformer-Based Speech Recognition On Extreme Edge-Computing Devices

With increasingly more powerful compute capabilities and resources in today's devices, traditionally compute-intensive automatic speech recognition (ASR) has been moving from the cloud to devices to better protect user privacy. However, it is still challenging to implement on-device ASR on resource-constrained devices, such as smartphones, smart wearables, and other smart home automation devices. In this paper, we propose a series of model architecture adaptions, neural network graph transformations, and numerical optimizations to fit an advanced Conformer based end-to-end streaming ASR system on resource-constrained devices without accuracy degradation. We achieve over 5.26 times faster than realtime (0.19 RTF) speech recognition on smart wearables while minimizing energy consumption and achieving state-of-the-art accuracy. The proposed methods are widely applicable to other transformer-based server-free AI applications. In addition, we provide a complete theory on optimal pre-normalizers that numerically stabilize layer normalization in any Lp-norm using any floating point precision.

cs.LG↗

arXiv2024

CT-MVSNet: Efficient Multi-View Stereo with Cross-scale Transformer

Recent deep multi-view stereo (MVS) methods have widely incorporated transformers into cascade network for high-resolution depth estimation, achieving impressive results. However, existing transformer-based methods are constrained by their computational costs, preventing their extension to finer stages. In this paper, we propose a novel cross-scale transformer (CT) that processes feature representations at different stages without additional computation. Specifically, we introduce an adaptive matching-aware transformer (AMT) that employs different interactive attention combinations at multiple scales. This combined strategy enables our network to capture intra-image context information and enhance inter-image feature relationships. Besides, we present a dual-feature guided aggregation (DFGA) that embeds the coarse global semantic information into the finer cost volume construction to further strengthen global and local feature awareness. Meanwhile, we design a feature metric loss (FM Loss) that evaluates the feature bias before and after transformation to reduce the impact of feature mismatch on depth estimation. Extensive experiments on DTU dataset and Tanks and Temples (T\&T) benchmark demonstrate that our method achieves state-of-the-art results. Code is available at https://github.com/wscstrive/CT-MVSNet.

cs.CV↗

arXiv2023

Understanding Resolution of Multi-Language Bugs: An Empirical Study on Apache Projects

Background: In modern software systems, more and more systems are written in multiple programming languages (PLs). There is no comprehensive investigation on the phenomenon of multi-programming-language (MPL) bugs, which resolution involves source files written in multiple PLs. Aim: This work investigated the characteristics of bug resolution in MPL software systems and explored the reasons why bug resolution involves multiple PLs. Method: We conducted an empirical study on 54 MPL projects selected from 655 Apache OSS projects, of which 66,932 bugs were analyzed. Results: (1) the percentage of MPL bugs (MPLBs) in the selected projects ranges from 0.17% to 42.26%, and the percentage of MPLBs for all projects as a whole is 10.01%; (2) 95.0% and 4.5% of all the MPLBs involve source files written in 2 and 3 PLs, respectively; (3) the change complexity resolution characteristics of MPLBs tend to be higher than those of single-programming-language bugs (SPLBs); (4) the open time for MPLBs is 19.52% to 529.57% significantly longer than SPLBs regarding 9 PL combinations; (5) the reopen rate of bugs involving the PL combination of JavaScript and Python reaches 20.66%; (6) we found 6 causes why the bug resolution involves multiple PLs and identified 5 cross-language calling mechanisms. Conclusion: MPLBs are related to increased development difficulty.

cs.SE↗