arXiv subjects

Dexin Wang

Publications and source records attributed to Dexin Wang.

At least 19 recordsLinked to original sources

arXiv2026

Dynamo: Dynamic Skill-Tool Evolution for Vision-Language Agents

Improving vision-language models (VLMs) on visual reasoning typically requires retraining or hand-designed prompts and tools. We present Dynamo, a training-free framework that adapts a frozen VLM without any weight updates. On a small labeled training subset, the agent inspects its own correct and incorrect attempts and evolves two complementary capabilities: reusable reasoning skills for cognitive bottlenecks, and executable visual tools for perceptual ones. Each generated tool is paired with a skill that specifies when to invoke it, and both capability types accumulate in a persistent library. Across four visual reasoning benchmarks and five VLM backbones, Dynamo improves direct inference on all 20 model--benchmark settings (avg. +5.6 acc). When the tool set is given in advance, the framework learns when to call each tool, and per-step tool choice improves on every tested backbone. Against task-specific RL (VTool-R1, DeepEyes), Dynamo closes 65--99% of the RL gap at a fraction of the compute, and combines additively with RL when available.

cs.AI↗

arXiv2026

Fill the GAP: A Granular Alignment Paradigm for Visual Reasoning in Multimodal Large Language Models

Visual latent reasoning lets a multimodal large language model (MLLM) create intermediate visual evidence as continuous tokens, avoiding external tools or image generators. However, existing methods usually follow an output-as-input latent paradigm and yield unstable gains. We identify evidence for a feature-space mismatch that can contribute to this instability: dominant visual-latent models build on pre-norm MLLMs and reuse decoder hidden states as predicted latent inputs, even though these states occupy a substantially different norm regime from the input embeddings the model was trained to consume (Xie et al., 2025; Li et al., 2026; Team et al., 2026). This mismatch can make direct latent feedback unreliable. Motivated by this diagnosis, we propose GAP, a Granular Alignment Paradigm for visual latent modeling. GAP aligns visual latent reasoning at three levels: feature-level alignment maps decoder outputs into input-compatible visual latents through a lightweight PCA-aligned latent head; context-level alignment grounds latent targets with inspectable auxiliary visual supervision; and capacity-guided alignment assigns latent supervision selectively to examples where the base MLLM struggles. On Qwen2.5-VL 7B, the resulting model achieves the best mean aggregate perception and reasoning performance among our supervised variants. Inference-time intervention probing further suggests that generated latents provide task-relevant visual signal beyond merely adding token slots.

cs.CV↗

arXiv2026

Impact of Background Dense Multipath Components on Multi-Band Fusion ISAC Systems

eess.SP↗

arXiv2026

Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills

Large Language Model (LLM) agents increasingly rely on domain-specific skills, yet manually authoring such skills does not scale, and skills generated purely from parametric knowledge often miss critical operational pitfalls. We introduce Trace2Skill, a framework that consolidates broad execution trajectories in parallel into a unified skill directory through inductive reasoning over agent experience. Trace2Skill supports both deepening existing human-written skills and creating useful skills from weak LLM-generated drafts. Experiments demonstrate the effectiveness of Trace2Skill across diverse domains, including office workflows, math reasoning, and vision QA. Importantly, the evolved skills are not merely memorized artifacts of the trajectories used to create them: they often transfer across model scales, across model families, and to out-of-distribution settings. For example, skills evolved from Qwen3.5-35B trajectories improve a Qwen3.5-122B agent by up to $57.65$ percentage points on WikiTableQuestions. Further analyses show that Trace2Skill outperforms sequential skill editing and ReasoningBank-style retrieval memories, compresses recurring failures and workarounds into standard operating procedures (SoPs), and yields portable skills that can be reused without parameter updates or test-time retrieval.

cs.AI↗

arXiv2026

Multi-Band Sensing in FR3 with Background Dense Multipath Components

Multi-band sensing has emerged as a key enabler of integrated sensing and communication (ISAC), one of the six primary usage scenarios defined for IMT-2030 (6G). The introduction of frequency range 3 (FR3, 7-24 GHz), comprising non-contiguous sub-bands across a wide frequency span, further reinforces the importance of multi-band operation. In such scenarios, frequency-dependent propagation effects that are collectively referred to as dense multipath components (DMC), including clutter, diffraction, and diffuse scattering, must be carefully considered. Building on prior literature and our experimental observations, this paper proposes a novel ISAC channel analysis tailored to multi-band sensing, based on a channel model with background DMCs. It also assesses the sensing trade-offs among sub-bands by analyzing Cram\'er-Rao bound (CRB)-based fundamental limits. Furthermore, a scalable multi-band estimator is proposed that resolves angular ambiguities arising from the grating lobes effect. Simulation results of the multi-band estimator demonstrate substantial gains in estimation accuracy and reductions in false alarm rate over single-band estimators operating on each constituent sub-band within the CRB-achieving regime. In a representative test case, the proposed estimator achieves reductions of 37.41% and 17.04% in the root mean squared error of delay estimation compared to single-band estimators operating at 8.75 GHz and 21.7 GHz, respectively.

eess.SP↗

arXiv2025

Compressed Multiband Sensing in FR3 Using Alternating Direction Method of Multipliers

Joint detection and localization of users and scatterers in multipath-rich channels on multiple bands is critical for integrated sensing and communication (ISAC) in 6G. Existing multiband sensing methods are limited by classical beamforming or computationally expensive approaches. This paper introduces alternating direction method of multipliers (ADMM)-assisted compressed multiband sensing (CMS), hereafter referred to as ADMM-CMS, which is a novel framework for multiband sensing using uplink quadrature amplitude modulation-modulated pilot symbols. To solve the CMS problem, we develop an adaptive ADMM algorithm that adjusts to noise and ensures automatic stopping if converged. ADMM combines the decomposability of dual ascent with the robustness of augmented Lagrangian methods, making it suitable for large-scale structured optimization. Simulations show that ADMM-CMS achieves higher spatial resolution and improved denoising compared to Bartlett-type beamforming, yielding a 34 dB gain in per-antenna transmit power for achieving a 0.9 successful recovery probability (SRP). Moreover, compared to performing compressed sensing separately on the constituent 7 GHz and 10 GHz sub-bands, ADMM-CMS achieves reductions in delay root mean squared error of 34.46% and 40.76%, respectively, at -41 dBm per-antenna transmit power, while also yielding improved SRP. Our findings demonstrate ADMM-CMS as an efficient enabler of ISAC in frequency range 3 (FR3, 7-24 GHz) for 6G systems.

eess.SP↗

arXiv2025

TextlessRAG: End-to-End Visual Document RAG by Speech Without Text

Document images encapsulate a wealth of knowledge, while the portability of spoken queries enables broader and flexible application scenarios. Yet, no prior work has explored knowledge base question answering over visual document images with queries provided directly in speech. We propose TextlessRAG, the first end-to-end framework for speech-based question answering over large-scale document images. Unlike prior methods, TextlessRAG eliminates ASR, TTS and OCR, directly interpreting speech, retrieving relevant visual knowledge, and generating answers in a fully textless pipeline. To further boost performance, we integrate a layout-aware reranking mechanism to refine retrieval. Experiments demonstrate substantial improvements in both efficiency and accuracy. To advance research in this direction, we also release the first bilingual speech--document RAG dataset, featuring Chinese and English voice queries paired with multimodal document content. Both the dataset and our pipeline will be made available at repository:https://github.com/xiepeijinhit-hue/textlessrag

cs.CV↗

arXiv2025

AdaST: Dynamically Adapting Encoder States in the Decoder for End-to-End Speech-to-Text Translation

In end-to-end speech translation, acoustic representations learned by the encoder are usually fixed and static, from the perspective of the decoder, which is not desirable for dealing with the cross-modal and cross-lingual challenge in speech translation. In this paper, we show the benefits of varying acoustic states according to decoder hidden states and propose an adaptive speech-to-text translation model that is able to dynamically adapt acoustic states in the decoder. We concatenate the acoustic state and target word embedding sequence and feed the concatenated sequence into subsequent blocks in the decoder. In order to model the deep interaction between acoustic states and target hidden states, a speech-text mixed attention sublayer is introduced to replace the conventional cross-attention network. Experiment results on two widely-used datasets show that the proposed method significantly outperforms state-of-the-art neural speech translation models.

cs.CL↗

arXiv2024

Goal State Generation for Robotic Manipulation Based on Linguistically Guided Hybrid Gaussian Diffusion

In robotic manipulation tasks, achieving a designated target state for the manipulated object is often essential to facilitate motion planning for robotic arms. Specifically, in tasks such as hanging a mug, the mug must be positioned within a feasible region around the hook. Previous approaches have enabled the generation of multiple feasible target states for mugs; however, these target states are typically generated randomly, lacking control over the specific generation locations. This limitation makes such methods less effective in scenarios where constraints exist, such as hooks already occupied by other mugs or when specific operational objectives must be met. Moreover, due to the frequent physical interactions between the mug and the rack in real-world hanging scenarios, imprecisely generated target states from end-to-end models often result in overlapping point clouds. This overlap adversely impacts subsequent motion planning for the robotic arm. To address these challenges, we propose a Linguistically Guided Hybrid Gaussian Diffusion (LHGD) network for generating manipulation target states, combined with a gravity coverage coefficient-based method for target state refinement. To evaluate our approach under a language-specified distribution setting, we collected multiple feasible target states for 10 types of mugs across 5 different racks with 10 distinct hooks. Additionally, we prepared five unseen mug designs for validation purposes. Experimental results demonstrate that our method achieves the highest success rates across single-mode, multi-mode, and language-specified distribution manipulation tasks. Furthermore, it significantly reduces point cloud overlap, directly producing collision-free target states and eliminating the need for additional obstacle avoidance operations by the robotic arm.

cs.RO↗

arXiv2024

Expand VSR Benchmark for VLLM to Expertize in Spatial Rules

Distinguishing spatial relations is a basic part of human cognition which requires fine-grained perception on cross-instance. Although benchmarks like MME, MMBench and SEED comprehensively have evaluated various capabilities which already include visual spatial reasoning(VSR). There is still a lack of sufficient quantity and quality evaluation and optimization datasets for Vision Large Language Models(VLLMs) specifically targeting visual positional reasoning. To handle this, we first diagnosed current VLLMs with the VSR dataset and proposed a unified test set. We found current VLLMs to exhibit a contradiction of over-sensitivity to language instructions and under-sensitivity to visual positional information. By expanding the original benchmark from two aspects of tunning data and model structure, we mitigated this phenomenon. To our knowledge, we expanded spatially positioned image data controllably using diffusion models for the first time and integrated original visual encoding(CLIP) with other 3 powerful visual encoders(SigLIP, SAM and DINO). After conducting combination experiments on scaling data and models, we obtained a VLLM VSR Expert(VSRE) that not only generalizes better to different instructions but also accurately distinguishes differences in visual positional information. VSRE achieved over a 27\% increase in accuracy on the VSR test set. It becomes a performant VLLM on the position reasoning of both the VSR dataset and relevant subsets of other evaluation benchmarks. We open-sourced the expanded model with data and Appendix at \url{https://github.com/peijin360/vsre} and hope it will accelerate advancements in VLLM on VSR learning.

cs.CV↗

arXiv2024

Hierarchical Diffusion Policy: manipulation trajectory generation via contact guidance

Decision-making in robotics using denoising diffusion processes has increasingly become a hot research topic, but end-to-end policies perform poorly in tasks with rich contact and have limited controllability. This paper proposes Hierarchical Diffusion Policy (HDP), a new imitation learning method of using objective contacts to guide the generation of robot trajectories. The policy is divided into two layers: the high-level policy predicts the contact for the robot's next object manipulation based on 3D information, while the low-level policy predicts the action sequence toward the high-level contact based on the latent variables of observation and contact. We represent both level policies as conditional denoising diffusion processes, and combine behavioral cloning and Q-learning to optimize the low level policy for accurately guiding actions towards contact. We benchmark Hierarchical Diffusion Policy across 6 different tasks and find that it significantly outperforms the existing state of-the-art imitation learning method Diffusion Policy with an average improvement of 20.8%. We find that contact guidance yields significant improvements, including superior performance, greater interpretability, and stronger controllability, especially on contact-rich tasks. To further unlock the potential of HDP, this paper proposes a set of key technical contributions including snapshot gradient optimization, 3D conditioning, and prompt guidance, which improve the policy's optimization efficiency, spatial awareness, and controllability respectively. Finally, real world experiments verify that HDP can handle both rigid and deformable objects.

cs.RO↗

arXiv2024

DRIP: A Versatile Family of Space-Time ISAC Discrete-time Sequences

The following paper introduces Dual beam-similarity awaRe Integrated sensing and communications (ISAC) with controlled Peak-to-average power ratio (DRIP) waveforms. DRIP is a novel family of space-time ISAC waveforms designed for dynamic peak-to-average power ratio (PAPR) adjustment. The proposed DRIP waveforms are designed to conform to specified PAPR levels while exhibiting beampattern properties, effectively targeting multiple desired directions and suppressing interference for multi-target sensing applications, while closely resembling radar chirps. For communication purposes, the proposed DRIP waveforms aim to minimize multi-user interference across various constellations. Addressing the non-convexity of the optimization framework required for generating DRIP waveforms, we introduce a block cyclic coordinate descent algorithm. This iterative approach ensures convergence to an optimal ISAC waveform solution. Simulation results validate the DRIP waveforms' superior performance, versatility, and favorable ISAC trade-offs, highlighting their potential in advanced multi-target sensing and communication systems.

eess.SP↗

arXiv2023

RIS-Enabled Integrated Sensing and Communication for 6G Systems

The following paper proposes a new target localization system design using an architecture based on reconfigurable intelligent surfaces (RISs) and passive radars (PRs) for integrated sensing and communications systems. The preamble of the communication signal is exploited in order to perform target sensing tasks, which involve detection and localization. The RIS in this case can aid the PR in sensing targets that are otherwise not seen by the PR itself, due to the many obstacles encountered within the propagation channel. Therefore, this work proposes a localization algorithm tailored for the integrated sensing and communications RIS-aided architecture, which is capable of uniquely positioning targets within the scene. The algorithm is capable of detecting the number of targets along with estimating the position of targets via angles and times of arrival. Our simulation results demonstrate the performance of the localization method in terms of different localization and detection metrics and for increasing RIS sizes.

eess.SP↗

arXiv2023

Synchrophasor Data Anomaly Detection on Grid Edge by 5G Communication and Adjacent Compute

The fifth-generation mobile communication (5G) technology offers opportunities to enhance the real-time monitoring of grids. The 5G-enabled phasor measurement units (PMUs) feature flexible positioning and cost-effective long-term maintenance without the constraints of fixing wires. This paper is the first to demonstrate the applicability of 5G in PMU communication, and the experiment was carried out at Verizon non-standalone test-bed at Pacific Northwest National Laboratory (PNNL) Advanced Wireless Communication lab. The performance of the 5G-enabled PMU communication setup is reviewed and discussed in this paper, and a generalized dynamic linear model (GDLM) based real-time synchrophasor data anomaly detection use-case is presented. Last but not least, the practicability of implementing 5G for wide-area protection strategies is explored and discussed by analyzing the experimental results.

eess.SY↗

arXiv2023

CCMB: A Large-scale Chinese Cross-modal Benchmark

Vision-language pre-training (VLP) on large-scale datasets has shown premier performance on various downstream tasks. In contrast to plenty of available benchmarks with English corpus, large-scale pre-training datasets and downstream datasets with Chinese corpus remain largely unexplored. In this work, we build a large-scale high-quality Chinese Cross-Modal Benchmark named CCMB for the research community, which contains the currently largest public pre-training dataset Zero and five human-annotated fine-tuning datasets for downstream tasks. Zero contains 250 million images paired with 750 million text descriptions, plus two of the five fine-tuning datasets are also currently the largest ones for Chinese cross-modal downstream tasks. Along with the CCMB, we also develop a VLP framework named R2D2, applying a pre-Ranking + Ranking strategy to learn powerful vision-language representations and a two-way distillation method (i.e., target-guided Distillation and feature-guided Distillation) to further enhance the learning capability. With the Zero and the R2D2 VLP framework, we achieve state-of-the-art performance on twelve downstream datasets from five broad categories of tasks including image-text retrieval, image-text matching, image caption, text-to-image generation, and zero-shot image classification. The datasets, models, and codes are available at https://github.com/yuxie11/R2D2

cs.CV↗

arXiv2023

Multi-Stage Reinforcement Learning for Non-Prehensile Manipulation

Manipulating objects without grasping them enables more complex tasks, known as non-prehensile manipulation. Most previous methods only learn one manipulation skill, such as reach or push, and cannot achieve flexible object manipulation.In this work, we introduce MRLM, a Multi-stage Reinforcement Learning approach for non-prehensile Manipulation of objects.MRLM divides the task into multiple stages according to the switching of object poses and contact points.At each stage, the policy takes the point cloud-based state-goal fusion representation as input, and proposes a spatially-continuous action that including the motion of the parallel gripper pose and opening width.To fully unlock the potential of MRLM, we propose a set of technical contributions including the state-goal fusion representation, spatially-reachable distance metric, and automatic buffer compaction.We evaluate MRLM on an Occluded Grasping task which aims to grasp the object in configurations that are initially occluded.Compared with the baselines, the proposed technical contributions improve the success rate by at least 40\% and maximum 100\%, and avoids falling into local optimum.Our method demonstrates strong generalization to unseen object with shapes outside the training distribution.Moreover, MRLM can be transferred to real world with zero-shot transfer, achieving a 95\% success rate.Code and videos can be found at https://sites.google.com/view/mrlm.

cs.RO↗

arXiv2023

On-Policy Pixel-Level Grasping Across the Gap Between Simulation and Reality

Grasp detection in cluttered scenes is a very challenging task for robots. Generating synthetic grasping data is a popular way to train and test grasp methods, as is Dex-net and GraspNet; yet, these methods generate training grasps on 3D synthetic object models, but evaluate at images or point clouds with different distributions, which reduces performance on real scenes due to sparse grasp labels and covariate shift. To solve existing problems, we propose a novel on-policy grasp detection method, which can train and test on the same distribution with dense pixel-level grasp labels generated on RGB-D images. A Parallel-Depth Grasp Generation (PDG-Generation) method is proposed to generate a parallel depth image through a new imaging model of projecting points in parallel; then this method generates multiple candidate grasps for each pixel and obtains robust grasps through flatness detection, force-closure metric and collision detection. Then, a large comprehensive Pixel-Level Grasp Pose Dataset (PLGP-Dataset) is constructed and released; distinguished with previous datasets with off-policy data and sparse grasp samples, this dataset is the first pixel-level grasp dataset, with the on-policy distribution where grasps are generated based on depth images. Lastly, we build and test a series of pixel-level grasp detection networks with a data augmentation process for imbalance training, which learn grasp poses in a decoupled manner on the input RGB-D images. Extensive experiments show that our on-policy grasp method can largely overcome the gap between simulation and reality, and achieves the state-of-the-art performance. Code and data are provided at https://github.com/liuchunsense/PLGP-Dataset.

cs.RO↗

arXiv2022

A 5G Enabled Adaptive Computing Workflow for Greener Power Grid

5G wireless technology can deliver higher data speeds, ultra low latency, more reliability, massive network capacity, increased availability, and a more uniform user experience to users. It brings additional power to help address the challenges brought by renewable integration and decarbonization. In this paper, a 5G enabled adaptive computing workflow has been presented that consists of various computing resources, such as 5G equipment, edge computing, cluster, Graphics processing unit (GPU) and cloud computing, with two examples showing technical feasibility for edge-grid-cloud interaction for power system real-time monitoring, security assessment, and forecasting. Benefiting from the high speed data transport and massive connection capability of 5G, the workflow shows its potential to seamlessly integrate various applications at distributed and/or centralized locations to build more complex and powerful functions, with better flexibility.

eess.SY↗