arXiv subjects

Kui Wu

Publications and source records attributed to Kui Wu.

At least 37 records · Page 2Linked to original sources

arXiv2025

TrackVLA++: Unleashing Reasoning and Memory Capabilities in VLA Models for Embodied Visual Tracking

Embodied Visual Tracking (EVT) is a fundamental ability that underpins practical applications, such as companion robots, guidance robots and service assistants, where continuously following moving targets is essential. Recent advances have enabled language-guided tracking in complex and unstructured scenes. However, existing approaches lack explicit spatial reasoning and effective temporal memory, causing failures under severe occlusions or in the presence of similar-looking distractors. To address these challenges, we present TrackVLA++, a novel Vision-Language-Action (VLA) model that enhances embodied visual tracking with two key modules, a spatial reasoning mechanism and a Target Identification Memory (TIM). The reasoning module introduces a Chain-of-Thought paradigm, termed Polar-CoT, which infers the target's relative position and encodes it as a compact polar-coordinate token for action prediction. Guided by these spatial priors, the TIM employs a gated update strategy to preserve long-horizon target memory, ensuring spatiotemporal consistency and mitigating target loss during extended occlusions. Extensive experiments show that TrackVLA++ achieves state-of-the-art performance on public benchmarks across both egocentric and multi-camera settings. On the challenging EVT-Bench DT split, TrackVLA++ surpasses the previous leading approach by 5.1 and 12, respectively. Furthermore, TrackVLA++ exhibits strong zero-shot generalization, enabling robust real-world tracking in dynamic and occluded scenarios.

cs.RO↗

arXiv2025

BC-ADMM: An Efficient Non-convex Constrained Optimizer with Robotic Applications

Non-convex constrained optimizations are ubiquitous in robotic applications such as multi-agent navigation, UAV trajectory optimization, and soft robot simulation. For this problem class, conventional optimizers suffer from small step sizes and slow convergence. We propose BC-ADMM, a variant of Alternating Direction Method of Multiplier (ADMM), that can solve a class of non-convex constrained optimizations with biconvex constraint relaxation. Our algorithm allows larger step sizes by breaking the problem into small-scale sub-problems that can be easily solved in parallel. We show that our method has both theoretical convergence speed guarantees and practical convergence guarantees in the asymptotic sense. Through numerical experiments in a row of four robotic applications, we show that BC-ADMM has faster convergence than conventional gradient descent and Newton's method in terms of wall clock time.

math.OC↗

arXiv2025

Efficient Differentiable Contact Model with Long-range Influence

With the maturation of differentiable physics, its role in various downstream applications: such as model predictive control, robotic design optimization, and neural PDE solvers, has become increasingly important. However, the derivative information provided by differentiable simulators can exhibit abrupt changes or vanish altogether, impeding the convergence of gradient-based optimizers. In this work, we demonstrate that such erratic gradient behavior is closely tied to the design of contact models. We further introduce a set of properties that a contact model must satisfy to ensure well-behaved gradient information. Lastly, we present a practical contact model for differentiable rigid-body simulators that satisfies all of these properties while maintaining computational efficiency. Our experiments show that, even from simple initializations, our contact model can discover complex, contact-rich control signals, enabling the successful execution of a range of downstream locomotion and manipulation tasks.

cs.RO↗

arXiv2025

Auto Hair Card Extraction for Smooth Hair with Differentiable Rendering

Hair cards remain a widely used representation for hair modeling in real-time applications, offering a practical trade-off between visual fidelity, memory usage, and performance. However, generating high-quality hair card models remains a challenging and labor-intensive task. This work presents an automated pipeline for converting strand-based hair models into hair card models with a limited number of cards and textures while preserving the hairstyle appearance. Our key idea is a novel differentiable representation where each strand is encoded as a projected 2D spline in the texture space, which enables efficient optimization with differentiable rendering and structured results respecting the hair geometry. Based on this representation, we develop a novel algorithm pipeline, where we first cluster hair strands into initial hair cards and project the strands into the texture space. We then conduct a two-stage optimization where our first stage optimizes the texture and geometry of each hair card separately, and after texture reduction, our second stage conducts joint optimization of all the cards for fine-tuning. Put together, our method is evaluated on a wide range of hairstyles, including straight, wavy, curly, and coily hairs. To better capture the appearance of short or coily hair, we additionally support hair cap and cross-card. Furthermore, our framework supports seamless LoD transitions via texture sharing, balancing texture memory efficiency and visual quality.

cs.GR↗

arXiv2025

Handle-based Mesh Deformation Guided By Vision Language Model

Mesh deformation is a fundamental tool in 3D content manipulation. Despite extensive prior research, existing approaches often suffer from low output quality, require significant manual tuning, or depend on data-intensive training. To address these limitations, we introduce a training-free, handle-based mesh deformation method. % Our core idea is to leverage a Vision-Language Model (VLM) to interpret and manipulate a handle-based interface through prompt engineering. We begin by applying cone singularity detection to identify a sparse set of potential handles. The VLM is then prompted to select both the deformable sub-parts of the mesh and the handles that best align with user instructions. Subsequently, we query the desired deformed positions of the selected handles in screen space. To reduce uncertainty inherent in VLM predictions, we aggregate the results from multiple camera views using a novel multi-view voting scheme. % Across a suite of benchmarks, our method produces deformations that align more closely with user intent, as measured by CLIP and GPTEval3D scores, while introducing low distortion -- quantified via membrane energy. In summary, our approach is training-free, highly automated, and consistently delivers high-quality mesh deformations.

cs.GR↗

arXiv2025

UnrealZoo: Enriching Photo-realistic Virtual Worlds for Embodied AI

We introduce UnrealZoo, a collection of over 100 photo-realistic 3D virtual worlds built on Unreal Engine, designed to reflect the complexity and variability of open-world environments. We also provide a rich variety of playable entities, including humans, animals, robots, and vehicles for embodied AI research. We extend UnrealCV with optimized APIs and tools for data collection, environment augmentation, distributed training, and benchmarking. These improvements achieve significant improvements in the efficiency of rendering and communication, enabling advanced applications such as multi-agent interactions. Our experimental evaluation across visual navigation and tracking tasks reveals two key insights: 1) environmental diversity provides substantial benefits for developing generalizable reinforcement learning (RL) agents, and 2) current embodied agents face persistent challenges in open-world scenarios, including navigation in unstructured terrain, adaptation to unseen morphologies, and managing latency in the close-loop control systems for interacting in highly dynamic objects. UnrealZoo thus serves as both a comprehensive testing ground and a pathway toward developing more capable embodied AI systems for real-world deployment.

cs.AI↗

arXiv2025

CCL-XCoT: An Efficient Cross-Lingual Knowledge Transfer Method for Mitigating Hallucination Generation

Multilingual Large Language Models(MLLMs) demonstrate strong generalization across languages, yet they remain prone to hallucinations, especially in low-resource languages, due to training data imbalances. These hallucinations, which include inaccurate or fabricated outputs, are particularly problematic in domain-specific generation tasks (Chataigner et al., 2024). To address this challenge, we propose CCL-XCoT(Curriculum-based Contrastive Learning-based Cross-lingual Chain-of-Thought), a two-stage fine-tuning framework for mitigating hallucination in MLLMs. Our approach first enhances cross-lingual semantic alignment through curriculum-based contrastive learning combined with next-token prediction during continued pre-training. Building on this foundation, we then introduce a cross-lingual Chain-of-Thought (XCoT) prompting strategy during instruction fine-tuning, which guides the model to reason in a high-resource language before generating answers in the target low-resource language. Experimental results show that CCL-XCoT reduces hallucination rates by up to 62% and substantially improves factual knowledge transfer across language pairs, without relying on external retrieval or multi-model ensembles.

cs.CL↗

arXiv2025

EmbRACE-3K: Embodied Reasoning and Action in Complex Environments

Recent advanced vision-language models(VLMs) have demonstrated strong performance on passive, offline image and video understanding tasks. However, their effectiveness in embodied settings, which require online interaction and active scene understanding remains limited. In such scenarios, an agent perceives the environment from a first-person perspective, with each action dynamically shaping subsequent observations. Even state-of-the-art models such as GPT-4o, Claude 3.5 Sonnet, and Gemini 2.5 Pro struggle in open-environment interactions, exhibiting clear limitations in spatial reasoning and long-horizon planning. To address this gap, we introduce EmRACE-3K, a dataset of over 3,000 language-guided tasks situated in diverse, photorealistic environments constructed using Unreal Engine and the UnrealCV-Zoo framework. The tasks encompass a wide range of embodied challenges, including navigation, object manipulation, and multi-stage goal execution. Each task unfolds as a multi-step trajectory, pairing first-person visual observations with high-level instructions, grounded actions, and natural language rationales that express the agent's intent at every step. Using EmRACE-3K, we establish a benchmark to evaluate the embodied reasoning capabilities of VLMs across three key dimensions: Exploration, Dynamic Spatial-Semantic Reasoning, and Multi-stage Goal Execution. In zero-shot settings, all models achieve success rates below 20%, underscoring the challenge posed by our benchmark and the current limitations of VLMs in interactive environments. To demonstrate the utility of EmRACE-3K, we further fine-tune Qwen2.5-VL-7B using supervised learning followed by reinforcement learning. This approach yields substantial improvements across all three challenge categories, highlighting the dataset's effectiveness in enabling the development of embodied reasoning capabilities.

cs.CV↗

arXiv2025

TrackVLA: Embodied Visual Tracking in the Wild

Embodied visual tracking is a fundamental skill in Embodied AI, enabling an agent to follow a specific target in dynamic environments using only egocentric vision. This task is inherently challenging as it requires both accurate target recognition and effective trajectory planning under conditions of severe occlusion and high scene dynamics. Existing approaches typically address this challenge through a modular separation of recognition and planning. In this work, we propose TrackVLA, a Vision-Language-Action (VLA) model that learns the synergy between object recognition and trajectory planning. Leveraging a shared LLM backbone, we employ a language modeling head for recognition and an anchor-based diffusion model for trajectory planning. To train TrackVLA, we construct an Embodied Visual Tracking Benchmark (EVT-Bench) and collect diverse difficulty levels of recognition samples, resulting in a dataset of 1.7 million samples. Through extensive experiments in both synthetic and real-world environments, TrackVLA demonstrates SOTA performance and strong generalizability. It significantly outperforms existing methods on public benchmarks in a zero-shot manner while remaining robust to high dynamics and occlusion in real-world scenarios at 10 FPS inference speed. Our project page is: https://pku-epic.github.io/TrackVLA-web.

cs.RO↗

arXiv2025

VLM Can Be a Good Assistant: Enhancing Embodied Visual Tracking with Self-Improving Vision-Language Models

We introduce a novel self-improving framework that enhances Embodied Visual Tracking (EVT) with Vision-Language Models (VLMs) to address the limitations of current active visual tracking systems in recovering from tracking failure. Our approach combines the off-the-shelf active tracking methods with VLMs' reasoning capabilities, deploying a fast visual policy for normal tracking and activating VLM reasoning only upon failure detection. The framework features a memory-augmented self-reflection mechanism that enables the VLM to progressively improve by learning from past experiences, effectively addressing VLMs' limitations in 3D spatial reasoning. Experimental results demonstrate significant performance improvements, with our framework boosting success rates by $72\%$ with state-of-the-art RL-based approaches and $220\%$ with PID-based methods in challenging environments. This work represents the first integration of VLM-based reasoning to assist EVT agents in proactive failure recovery, offering substantial advances for real-world robotic applications that require continuous target monitoring in dynamic, unstructured environments. Project website: https://sites.google.com/view/evt-recovery-assistant.

cs.CV↗

arXiv2025

Hierarchical Instruction-aware Embodied Visual Tracking

User-Centric Embodied Visual Tracking (UC-EVT) presents a novel challenge for reinforcement learning-based models due to the substantial gap between high-level user instructions and low-level agent actions. While recent advancements in language models (e.g., LLMs, VLMs, VLAs) have improved instruction comprehension, these models face critical limitations in either inference speed (LLMs, VLMs) or generalizability (VLAs) for UC-EVT tasks. To address these challenges, we propose \textbf{Hierarchical Instruction-aware Embodied Visual Tracking (HIEVT)} agent, which bridges instruction comprehension and action generation using \textit{spatial goals} as intermediaries. HIEVT first introduces \textit{LLM-based Semantic-Spatial Goal Aligner} to translate diverse human instructions into spatial goals that directly annotate the desired spatial position. Then the \textit{RL-based Adaptive Goal-Aligned Policy}, a general offline policy, enables the tracker to position the target as specified by the spatial goal. To benchmark UC-EVT tasks, we collect over ten million trajectories for training and evaluate across one seen environment and nine unseen challenging environments. Extensive experiments and real-world deployments demonstrate the robustness and generalizability of HIEVT across diverse environments, varying target dynamics, and complex instruction combinations. The complete project is available at https://sites.google.com/view/hievt.

cs.CV↗

arXiv2025

CALM: Co-evolution of Algorithms and Language Model for Automatic Heuristic Design

Tackling complex optimization problems often relies on expert-designed heuristics, typically crafted through extensive trial and error. Recent advances demonstrate that large language models (LLMs), when integrated into well-designed evolutionary search frameworks, can autonomously discover high-performing heuristics at a fraction of the traditional cost. However, existing approaches predominantly rely on verbal guidance, i.e., manipulating the prompt generation process, to steer the evolution of heuristics, without adapting the underlying LLM. We propose a hybrid framework that combines verbal and numerical guidance, the latter achieved by fine-tuning the LLM via reinforcement learning based on the quality of generated heuristics. This joint optimization allows the LLM to co-evolve with the search process. Our method outperforms state-of-the-art (SOTA) baselines across various optimization tasks, running locally on a single 24GB GPU using a 7B model with INT4 quantization. It surpasses methods that rely solely on verbal guidance, even when those use significantly more powerful API-based models.

cs.NE↗

arXiv2024

SDRS: Shape-Differentiable Robot Simulator

Robot simulators are indispensable tools across many fields, and recent research has significantly improved their functionality by incorporating additional gradient information. However, existing differentiable robot simulators suffer from non-differentiable singularities, when robots undergo substantial shape changes. To address this, we present the Shape-Differentiable Robot Simulator (SDRS), designed to be differentiable under significant robot shape changes. The core innovation of SDRS lies in its representation of robot shapes using a set of convex polyhedrons. This approach allows us to generalize smooth, penalty-based contact mechanics for interactions between any pair of convex polyhedrons. Using the separating hyperplane theorem, SDRS introduces a separating plane for each pair of contacting convex polyhedrons. This separating plane functions as a zero-mass auxiliary entity, with its state determined by the principle of least action. This setup ensures global differentiability, even as robot shapes undergo significant geometric and topological changes. To demonstrate the practical value of SDRS, we provide examples of robot co-design scenarios, where both robot shapes and control movements are optimized simultaneously.

cs.RO↗

arXiv2024

Decoupling of carbonate-organic carbon isotope during the Carnian Pluvial Episode

The Carnian Pluvial Episode (CPE) was a major global climate change event in the early Late Triassic that significantly affected marine ecosystems and carbon cycles. One of the most prominent features of the CPE is the coupled multiple negative carbonate-organic carbon isotope excursions. However, at Erguan and Xiashulao from eastern Tethys, a decoupling between carbonate-organic carbon isotope during CPE was observed. At the end of early Carnian (Julian), the carbonate carbon isotope showed a negative excursion of 2-3 per-mille, while the organic carbon isotope exhibited a positive excursion of about 3-4 per-mille. In addition, increased terrestrial inputs is indicated by the rising C/N (3 to 10) and decreasing Y/Ho (42 to 27) that coexist with this decoupling. The coupling of carbon isotope negative excursions is from the shallow shelves and the deep slopes, whereas the decoupling occurs from the deep shelf to the shallow slope. In the deep shelf to the shallow slope, sedimentary organic matter is mainly sourced from pelagic before the CPE as evidenced by low C/N (3) and high Y/Ho (36-42). During the CPE, the increased fresh water flux (Sr/Ba <1) enhanced terrestrial input in organic matter, which may cause positive excursions in the carbon isotope record with elevated TOC content. As a result, the carbonate-organic carbon isotope decoupled. In contrast, organic matter in sediments from the shallow shelf and deep slope are mainly from terrestrial and pelagic sources, respectively. This study reveals the significant impact of terrestrial inputs on marine carbon cycling during the Carnian Pluvial Episode, highlighting the crucial role of climate events in modifying the carbon isotope record.

physics.ao-ph↗

arXiv2024

XFormParser: A Simple and Effective Multimodal Multilingual Semi-structured Form Parser

In the domain of Document AI, parsing semi-structured image form is a crucial Key Information Extraction (KIE) task. The advent of pre-trained multimodal models significantly empowers Document AI frameworks to extract key information from form documents in different formats such as PDF, Word, and images. Nonetheless, form parsing is still encumbered by notable challenges like subpar capabilities in multilingual parsing and diminished recall in industrial contexts in rich text and rich visuals. In this work, we introduce a simple but effective \textbf{M}ultimodal and \textbf{M}ultilingual semi-structured \textbf{FORM} \textbf{PARSER} (\textbf{XFormParser}), which anchored on a comprehensive Transformer-based pre-trained language model and innovatively amalgamates semantic entity recognition (SER) and relation extraction (RE) into a unified framework. Combined with Bi-LSTM, the performance of multilingual parsing is significantly improved. Furthermore, we develop InDFormSFT, a pioneering supervised fine-tuning (SFT) industrial dataset that specifically addresses the parsing needs of forms in various industrial contexts. XFormParser has demonstrated its unparalleled effectiveness and robustness through rigorous testing on established benchmarks. Compared to existing state-of-the-art (SOTA) models, XFormParser notably achieves up to 1.79\% F1 score improvement on RE tasks in language-specific settings. It also exhibits exceptional cross-task performance improvements in multilingual and zero-shot settings. The codes, datasets, and pre-trained models are publicly available at https://github.com/zhbuaa0/xformparser.

cs.CL↗

arXiv2024

BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction

Simulating realistic behaviors of traffic agents is pivotal for efficiently validating the safety of autonomous driving systems. Existing data-driven simulators primarily use an encoder-decoder architecture to encode the historical trajectories before decoding the future. However, the heterogeneity between encoders and decoders complicates the models, and the manual separation of historical and future trajectories leads to low data utilization. Given these limitations, we propose BehaviorGPT, a homogeneous and fully autoregressive Transformer designed to simulate the sequential behavior of multiple agents. Crucially, our approach discards the traditional separation between "history" and "future" by modeling each time step as the "current" one for motion generation, leading to a simpler, more parameter- and data-efficient agent simulator. We further introduce the Next-Patch Prediction Paradigm (NP3) to mitigate the negative effects of autoregressive modeling, in which models are trained to reason at the patch level of trajectories and capture long-range spatial-temporal interactions. Despite having merely 3M model parameters, BehaviorGPT won first place in the 2024 Waymo Open Sim Agents Challenge with a realism score of 0.7473 and a minADE score of 1.4147, demonstrating its exceptional performance in traffic agent simulation.

cs.AI↗

arXiv2024

Real-time Level-of-Detail Strand-based Hair Rendering

Strand-based hair rendering has become increasingly popular in production for its realistic appearance. However, the prevailing level-of-detail solution employing hair cards for distant hair models introduces a significant discontinuity in dynamics and appearance during the transition from strands to cards. We introduce an innovative real-time framework for strand-based hair rendering that ensures seamless transitions between different levels of detail (LOD) while maintaining a consistent hair appearance. Our method uses elliptical thick hairs that contain multiple hair strands at each LOD to maintain the shapes of hair clusters. In addition to geometric fitting, we formulate an elliptical Bidirectional Curve Scattering Distribution Functions (BCSDF) model for a thick hair, accurately capturing single scattering and multiple scattering within the hair cluster, accommodating a spectrum from sparse to dense hair distributions. Our framework, tested on various hairstyles with dynamics as well as knits, shows that it can produce highly similar appearances to full hair geometries at different viewing distances with seamless LOD transitions, while achieving up to a 3x speedup.

cs.GR↗

arXiv2024

Volumetric Homogenization for Knitwear Simulation

This paper presents volumetric homogenization, a spatially varying homogenization scheme for knitwear simulation. We are motivated by the observation that macro-scale fabric dynamics is strongly correlated with its underlying knitting patterns. Therefore, homogenization towards a single material is less effective when the knitting is complex and non-repetitive. Our method tackles this challenge by homogenizing the yarn-level material locally at volumetric elements. Assigning a virtual volume of a knitting structure enables us to model bending and twisting effects via a simple volume-preserving penalty and thus effectively alleviates the material nonlinearity. We employ an adjoint Gauss-Newton formulation to battle the dimensionality challenge of such per-element material optimization. This intuitive material model makes the forward simulation GPU-friendly. To this end, our pipeline also equips a novel domain-decomposed subspace solver crafted for GPU projective dynamics, which makes our simulator hundreds of times faster than the yarn-level simulator. Experiments validate the capability and effectiveness of volumetric homogenization. Our method produces realistic animations of knitwear matching the quality of full-scale yarn-level simulations. It is also orders of magnitude faster than existing homogenization techniques in both the training and simulation stages.

cs.GR↗