论文检索

Liang Qin, Min Wang, Xingyu Lu, Aowen Qiu, Wengang Zhou, Houqiang Li

Active search and tracking of arbitrary targets by Unmanned Aerial Vehicles (UAVs) in cluttered environments remains a highly challenging problem. Existing methods either construct complex modular pipelines, leading to substantial computational costs, or adopt end-to-end controllers that often fail to generalize across different targets and scenes. Moreover, search and tracking are typically treated separately despite their strong interdependence. In this paper, we present UAST, a simple yet effective mapping-free framework that unifies active search and persistent tracking using only RGB-D observations. The proposed system couples a dual-branch perception module with a Rule-Based Point Search Policy that adaptively switches between tracking and search-based recovery. A lightweight control network generates dynamically feasible trajectories directly from fused perception and UAV states. Furthermore, we introduce a training strategy with an elaborated tracking-aware visibility loss and a tailored data construction. Extensive experiments in both simulated and real-world environments show that our approach achieves higher success rates, more stable long-term tracking, and faster target search compared with existing methods, while maintaining high efficiency. The code is available at https://github.com/qinliangql/UAST.

Yuqiu Liu, Jialin Song, Marissa Ramirez de Chanlatte, Rochishnu Chowdhury, Rushil Paresh Desai, Wuyang Chen, Daniel Martin, Michael W. Mahoney

Real objects that inhabit the physical world follow physical laws and thus behave plausibly during interaction with other physical objects. However, current methods that perform 3D reconstructions of real-world scenes from multi-view 2D images optimize primarily for visual fidelity, i.e., they train with photometric losses and reason about uncertainty in the image or representation space. This appearance-centric view overlooks body contacts and couplings, conflates function-critical regions (e.g., aerodynamic or hydrodynamic surfaces) with ornamentation, and reconstructs structures suboptimally, even when physical regularizers are added. All these can lead to unphysical and implausible interactions. To address this, we consider the question: How can 3D reconstruction become aware of real-world interactions and underlying object functionality, beyond visual cues? To answer this question, we propose FluidGaussian, a plug-and-play method that tightly couples geometry reconstruction with ubiquitous fluid-structure interactions to assess surface quality at high granularity. We define a simulation-based uncertainty metric induced by fluid simulations and integrate it with active learning to prioritize views that improve both visual and physical fidelity. In an empirical evaluation on NeRF Synthetic (Blender), Mip-NeRF 360, and DrivAerNet++, our FluidGaussian method yields up to +8.6% visual PSNR (Peak Signal-to-Noise Ratio) and -62.3% velocity divergence during fluid simulations. Our code is available in https://github.com/delta-lab-ai/FluidGaussian.

Shih-Wen Liu, Yen-Chang Chen, Wei-Ta Chu, Fu-En Yang, Yu-Chiang Frank Wang

Multi-task learning (MTL) aims to enable a single model to solve multiple tasks efficiently; however, current parameter-efficient fine-tuning (PEFT) methods remain largely limited to single-task adaptation. We introduce Free Sinewich, a parameter-efficient multi-task learning framework that enables near-zero-cost weight modulation via frequency switching (Free). Specifically, a Sine-AWB (Sinewich) layer combines low-rank factors and convolutional priors into a single kernel, which is then modulated elementwise by a sinusoidal transformation to produce task-specialized weights. A lightweight Clock Net is introduced to produce bounded frequencies that stabilize this modulation during training. Theoretically, sine modulation enhances the rank of low-rank adapters, while frequency separation decorrelates the weights of different tasks. On dense prediction benchmarks, Free Sinewich achieves state-of-the-art performance-efficiency trade-offs (e.g., up to +5.39% improvement over single-task fine-tuning with only 6.53M trainable parameters), offering a compact and scalable paradigm based on frequency-based parameter sharing.

Bukun Huang, Benlei Cui, Zhizeng Ye, Xuemei Dong, Tuo Chen, Hui Xue, Dingkang Yang, Longtao Huang, Haiwen Hong, Jingqun Tang

Text-to-image (T2I) diffusion models currently lack an efficient mechanism for early quality assessment, forcing costly random trial-and-error in scenarios requiring multiple generations (e.g., iterating on prompts, agent-based image generation, flow-grpo). To address this, we first reveal a strong correlation between the attention distribution in the early diffusion process and the final image quality. Building upon this insight, we introduce Diffusion Probe, a pioneering framework that leverages the model's internal cross-attention maps as a predictive signal. We propose a lightweight predictor, trained to establish a direct mapping from statistical properties of these nascent cross-attention distributions--extracted from the initial denoising steps--to the final image's comprehensive quality. This allows our probe to accurately forecast various aspects of image quality, regardless of the specific ground-truth quality metric, long before full synthesis is complete. We empirically validate the reliability and generalizability of Diffusion Probe through its consistently strong predictive accuracy across a wide spectrum of conditions. On diverse T2I models (e.g., SDXL, FLUX, Qwen-Image), throughout broad early-denoising windows, across various resolutions, and with different quality metrics, it achieves high correlation (PCC > 0.7) and classification performance (AUC-ROC > 0.9). This intrinsic reliability is further demonstrated in practice by successfully optimizing T2I workflows that benefit from early, quality-guided decisions, such as Prompt Optimization, Seed Selection, and Accelerated RL Training. In these applications, the probe's early signal enables more targeted sampling strategies, preempting costly computations on low-potential paths. This yields a dual benefit: a significant reduction in computational overhead and a simultaneous improvement in final outcome quality, establishing Diffusion Probe as a model-agnostic and broadly applicable tool poised to revolutionize T2I efficiency.

Muyan Zhong, Erfei Cui, Sen Xing, Weiyun Wang, Wen Wu, Yuchen Hu, Yanting Zhang, Xiaowei Hu, Wenhai Wang, Chao Zhang 等

Multimodal large language models (MLLMs) have expanded from vision-language systems to include audio, unlocking new capabilities in cross-modal reasoning and interaction. To address the limitation that existing benchmarks focus mainly on perception tasks and lack a unified cognitive evaluation framework, we propose Hierarchical Audio-Visual Evaluation Benchmark (HAVE-Bench). It systematically evaluates the audio-related capabilities of MLLMs along a three-level cognitive hierarchy: Perception, Reasoning, and Interaction, utilizing 2,451 curated samples and manually annotated multi-turn interaction-level tasks. Experiments using this unified framework reveal significant gaps in existing models at the reasoning and interaction levels, with speech-driven visual question answering (VQA) performance significantly lagging behind the text-image setting. These findings underscore the urgency of enhancing models' handling of long and complex audio and facilitating the transfer of reasoning capabilities from the vision-text to the audio-visual domain.

Arkaprava Sinha, Monish Soundar Raj, Pu Wang, Ahmed Helmy, Hieu Le, Srijan Das

Temporal Action Detection (TAD) in untrimmed videos poses significant challenges, particularly for Activities of Daily Living (ADL) requiring models to (1) process long-duration videos, (2) capture temporal variations in actions, and (3) simultaneously detect dense overlapping actions. Existing CNN and Transformer-based approaches, struggle to jointly capture fine-grained detail and long-range structure at scale. State-space Model (SSM) based Mamba offers powerful long-range modeling, but naive application to TAD collapses fine-grained temporal structure and fails to account for the challenges inherent to TAD. To this end, we propose Multi-Scale Temporal Mamba (MS-Temba), which extends Mamba to TAD with newly introduced dilated SSMs. Each Temba block, comprising dilated SSMs coupled with our proposed additional losses, enables the learning of discriminative representations across temporal scales. A lightweight Multi-scale Mamba Fuser then unifies these multi-scale features via SSM-based aggregation, yielding precise action-boundary localization. With only 17M parameters, MS-Temba achieves state-of-the-art performance on densely labeled ADL benchmarks TSU & Charades, and further generalizes to long-form video summarization, setting new state-of-the-art results on TVSum & SumMe.

Seungjae Lee, Yoonkyo Jung, Inkook Chun, Yao-Chih Lee, Zikui Cai, Hongjia Huang, Aayush Talreja, Tan Dao, Yongyuan Liang, Jia-Bin Huang 等

Learning new robot tasks on new platforms and in new scenes from only a handful of demonstrations remains challenging. While videos of other embodiments---humans and different robots---are abundant, differences in embodiment, camera, and environment hinder their direct use. We address the small-data problem by introducing a unifying, symbolic representation---a compact 3D "trace-space" of scene-level trajectories---that enables learning from cross-embodiment, cross-environment, and cross-task videos. We present TraceGen, a world model that predicts future motion in trace-space rather than pixel space, abstracting away appearance while retaining the geometric structure needed for manipulation. To train TraceGen at scale, we develop TraceForge, a data pipeline that transforms heterogeneous human and robot videos into consistent 3D traces, yielding a corpus of 123K videos and 1.8M observation--trace--language triplets. Pretraining on this corpus produces a transferable 3D motion prior that adapts efficiently: with just five target robot videos, TraceGen attains 80% success across four tasks while offering 50-600x faster inference than state-of-the-art video-based world models. In the more challenging case where only five uncalibrated human demonstration videos captured on a handheld phone are available, it still reaches 67.5% success on a real robot, highlighting TraceGen's ability to adapt across embodiments without heavy pixel-space generation.

Farhat Shaikh, Ayan Banerjee, Sandeep Gupta

We introduce EMMA, a physics-informed multimodal framework that recovers all identifiable dynamical parameters of a system directly from raw video, audio, and image-based time-series observations. Unlike prior video-only approaches that struggle with occluded states, hidden actuation inputs, or assumptions about known initial conditions and coordinate frames, EMMA performs joint inference of explicit parameters, implicit dynamical components, and calibration invariants within a unified continuous-time model. EMMA leverages a Liquid Time-Constant (LTC) network to learn latent dynamics from heterogeneous modalities while a physics-constrained loss enforces consistency with the governing differential equations. A unified feature pipeline enables consistent alignment across video trajectories, acoustic signatures, and chart-derived measurements, allowing EMMA to estimate parameters under forced, implicit, and multivariate dynamics without requiring segmentation masks, differentiable rendering, or specialized sensors. Across 100+ scenarios including five standard dynamical benchmarks (75 Delfys videos), real-world rover and quadrotor systems with hidden inputs, and simulation-chart case studies spanning biological and chaotic systems, EMMA delivers robust multi-parameter recovery and significantly outperforms existing single-modality and equation-discovery baselines. Our results establish EMMA as a general, scalable solution for physics-consistent model extraction from opportunistic multimodal data.

Tianyue Wang, Leigang Qu, Tianyu Yang, Xiangzhao Hao, Yifan Xu, Haiyun Guo, Jinqiao Wang

Zero-Shot Composed Image Retrieval (ZS-CIR) aims to retrieve target images given a multimodal query (comprising a reference image and a modification text), without training on annotated triplets. Existing methods typically convert the multimodal query into a single modality--either as an edited caption for Text-to-Image retrieval (T2I) or as an edited image for Image-to-Image retrieval (I2I). However, T2I often loses fine-grained visual details, while I2I struggles with complex semantic modifications. To effectively leverage their complementary strengths under diverse query intents, we propose WISER, a training-free framework that unifies T2I and I2I via a "retrieve-verify-refine" pipeline, explicitly modeling intent awareness and uncertainty awareness. Specifically, WISER first performs Wider Search by generating both edited captions and images for parallel retrieval to broaden the candidate pool. Then, it conducts Adaptive Fusion with a verifier to assess retrieval confidence, triggering refinement for uncertain retrievals, and dynamically fusing the dual-path for reliable ones. For uncertain retrievals, WISER generates refinement suggestions through structured self-reflection to guide the next retrieval round toward Deeper Thinking. Extensive experiments show the superior performance of WISER across multiple benchmarks, achieving relative improvements of 45% on CIRCO (mAP@5) and 57% on CIRR (Recall@1) over existing training-free methods. Code is released at https://github.com/Physicsmile/WISER.

Leijie Wang, Otilia Stretcu, Wei Qiao, Thomas Denby, Krishnamurthy Viswanathan, Enming Luo, Chun-Ta Lu, Tushar Dogra, Ranjay Krishna, Ariel Fuxman

From content moderation to content curation, applications requiring vision classifiers for visual concepts are rapidly expanding. Existing human-in-the-loop approaches typically assume users begin with a clear, stable concept understanding to be able to provide high-quality supervision. In reality, users often start with a vague idea and must iteratively refine it through "concept deliberation", a practice we uncovered through structured interviews with content moderation experts. We operationalize the common strategies in deliberation used by real content moderators into a human-in-the-loop framework called Agile Deliberation that explicitly supports evolving and subjective concepts. The system supports users in defining the concept for themselves by exposing them to borderline cases. The system does this with two deliberation stages: (1) concept scoping, which decomposes the initial concept into a structured hierarchy of sub-concepts, and (2) concept iteration, which surfaces semantically borderline examples for user reflection and feedback to iteratively align an image classifier with the user's evolving intent. Since concept deliberation is inherently subjective and interactive, we painstakingly evaluate the framework through 18 user sessions, each 1.5h long, rather than standard benchmarking datasets. We find that Agile Deliberation achieves 7.5% higher F1 scores than automated decomposition baselines and more than 3% higher than manual deliberation, while participants reported clearer conceptual understanding and lower cognitive effort.

Xiangyang Luo, Qingyu Li, Yuming Li, Guanbo Huang, Yongjie Zhu, Wenyu Qin, Meng Wang, Pengfei Wan, Shao-Lun Huang

Recent advances in video generation models have achieved impressive results. However, these models heavily rely on the use of high-quality data that combines both high visual quality and high motion quality. In this paper, we identify a key challenge in video data curation: the Motion-Vision Quality Dilemma. We discovered that visual quality and motion intensity inherently exhibit a negative correlation, making it hard to obtain golden data that excels in both aspects. To address this challenge, we first examine the hierarchical learning dynamics of video diffusion models and conduct gradient-based analysis on quality-degraded samples. We discover that quality-imbalanced data can produce gradients similar to golden data at appropriate timesteps. Based on this, we introduce the novel concept of Timestep selection in Training Process. We propose Timestep-aware Quality Decoupling (TQD), which modifies the data sampling distribution to better match the model's learning process. For certain types of data, the sampling distribution is skewed toward higher timesteps for motion-rich data, while high visual quality data is more likely to be sampled during lower timesteps. Through extensive experiments, we demonstrate that TQD enables training exclusively on separated imbalanced data to achieve performance surpassing conventional training with better data, challenging the necessity of perfect data in video generation. Moreover, our method also boosts model performance when trained on high-quality data, showcasing its effectiveness across different data scenarios.

Abdullah Azeem, Ruisheng Wang, Qingquan Li, Abubakar Siddique

Autonomous object detection in remote sensing requires systems that can discover new categories and assign them usable labels during deployment. Existing Open-World Object Detectors identify unknown objects but leave them unnamed until manual annotation. In contrast, Open-Vocabulary Detectors recognize unseen categories only with provided prompts at test time, lacking autonomous discovery or naming. This work presents HSGDet, a detector that achieves both discovery and semantic assignment at test time without external prompts. This method introduces DHGA that navigates a hierarchical semantic graph to perform scene-conditioned coarse-to-fine classification of detected objects. It leverages spatial co-occurrence patterns from surrounding scene context to produce classification confidence scores. High-scoring regions are identified as known objects, while low-scoring regions are flagged as unknown detections. Unknown regions pass to CR2T, which synthesizes text embeddings by fusing visual features, hierarchical parents, and scene context, enabling prompt-free labeling and vocabulary expansion. This approach enables prompt-free semantic labeling and supports autonomous vocabulary expansion without requiring external models. Results demonstrate that HSGDet outperforms state-of-the-art methods by a large margin of 6.6 points in Known mAP and 9.9 points in Unknown Recall. It also reduces Wilderness Impact by 36%, enabling scalable and autonomous aerial monitoring.

Jiuyang Dong, Jiahan Li, Junjun Jiang, Yongbing Zhang

Whole-slide images (WSIs) in computational pathology contain billions of pixels, making end-to-end training of feature extractors and multi-instance learning (MIL) networks infeasible on a single commodity GPU.Existing methods often freeze the feature extractor and train MIL networks on the resulting frozen features, which introduces a semantic gap that limits downstream performance. To address this issue, we propose FBTA, a Feature Bridging and Translation Alignment framework for WSI classification. FBTA is the first end-to-end MIL framework trainable on a single 24\,GB GPU, leveraging three complementary feature-bag views: end-to-end features enable joint optimization, frozen features stabilize training, and translated features support practical inference.Experiments on diverse datasets, including TCGA-NSCLC (Shot20/50/100) and TCGA-STAD, demonstrate the effectiveness and generality of FBTA, which consistently improves performance across three MIL architectures and two extractors. For example, with ResNet-50 as the extractor, FBTA improves the accuracy of the classic ABMIL by 13.1% and 15.8% on the NSCLC-Shot50 and TCGA-STAD datasets, respectively, and further enhances the state-of-the-art MambaMIL by 4.1% and 9.2% on the same datasets. Moreover, FBTA yields additional gains for MIL models that incorporate self-supervised pretraining strategies and data augmentation techniques.These results suggest FBTA is a feasible and scalable framework for end-to-end MIL on gigapixel WSIs. The code will be available.

Chenzhuo Liao, Xin Chen, Bingchen Li, Yu Meng, Tao Yue, Xuemei Hu

Reconstructing High Dynamic Range (HDR) video from alternately exposed Low Dynamic Range (LDR) frames is challenged by large motion, exposure-induced photometric inconsistency, and information loss in saturated or under-exposed regions. Prior HDR video pipelines typically follow an alignment-reconstruction paradigm, which is limited by the precision of alignment and the performance of the fusion module. We propose a new reconstruction framework called Learning Representation-enhanced HDR Video Reconstruction (LRHDR), which is built around two novel components: an Amalgamated Cross-exposure Consistent Representation (ACCR) network and an Adaptive Pixel-wise Sparse Weighted Fusion (APSWF). The ACCR includes an Exposure-aware Interleaved Context (EIC) encoder and a Representation Mapper (RM). The EIC couples a large-field path with a high-fidelity sub-pixel path and an exposure gate to produce exposure-aware features. The RM avoids explicit cross-exposure alignment by mapping features from different exposures into a unified representation via per-pixel, per-channel linear modulation and decoding into the calibrated linear HDR domain. The APSWF treats fusion as pixel-wise candidate selection, producing sparse weighted masks to form a normalized fusion in the linear HDR domain, thereby suppressing artifacts. Extensive experiments on standard benchmarks demonstrate that our LRHDR outperforms previous methods.

Jie Wang, Guang Li, Zhijian Huang, Chenxu Dang, Hangjun Ye, Yahong Han, Long Chen

The significance of cross-view 3D geometric modeling capabilities for autonomous driving is self-evident, yet existing Vision-Language Models (VLMs) inherently lack this capability, resulting in their mediocre performance. While some promising approaches attempt to mitigate this by constructing Q&A data for auxiliary training, they still fail to fundamentally equip VLMs with the ability to comprehensively handle diverse evaluation protocols. We thus chart a new course, advocating for the infusion of VLMs with the cross-view geometric grounding of mature 3D foundation models, closing this critical capability gap in autonomous driving. In this spirit, we propose a novel architecture, VGGDrive, which empowers Vision-language models with cross-view Geometric Grounding for autonomous Driving. Concretely, to bridge the cross-view 3D geometric features from the frozen visual 3D model with the VLM's 2D visual features, we introduce a plug-and-play Cross-View 3D Geometric Enabler (CVGE). The CVGE decouples the base VLM architecture and effectively empowers the VLM with 3D features through a hierarchical adaptive injection mechanism. Extensive experiments show that VGGDrive enhances base VLM performance across five autonomous driving benchmarks, including tasks like cross-view risk perception, motion prediction, and trajectory planning. It's our belief that mature 3D foundation models can empower autonomous driving tasks through effective integration, and we hope our initial exploration demonstrates the potential of this paradigm to the autonomous driving community.

Hakan Emre Gedik, Shashank Gupta, Alan Bovik

No-reference image quality assessment (NR IQA) has recently benefited from deep and multimodal models, yet many SOTA systems still violate at least one basic requirement: they either discard critical quality cues via aggressive resizing, fail to generalize across resolutions, cannot be jointly trained on heterogeneous IQA datasets with mismatched MOS scales, or require prohibitive computation. We present ReLIQS, a model for Resolution-agnostic Learning for Image Quality with Saliency, which is resolution-agnostic, preserves original-resolution quality cues, learns from multiple subjective studies, and remains computationally efficient and budget-adaptive. ReLIQS is a CLIP-based multiscale patch-driven architecture that learns both where to look and how to judge quality. Fixed-size patches are sampled across multiple resolutions, including the original resolution, and encoded with a CLIP vision backbone. A lightweight Perceptual Importance Estimator then predicts IQA-specific importance maps to select a small set of informative patches, and a Quality Aspect Module aggregates their embeddings into a single image-level score. Across authentic, synthetic, and AIGC benchmarks spanning diverse resolutions and distortions, ReLIQS generalizes better than strong CNN-, CLIP-, and MLLM-based baselines with matching or reduced computational cost.

Yuanhang Lei, Tao Cheng, Xingxuan Li, Boming Zhao, Siyuan Huang, Ruizhen Hu, Peter Yichen Chen, Hujun Bao, Zhaopeng Cui

Achieving real-time physics-based animation that generalizes across diverse 3D shapes and discretizations remains a fundamental challenge. We introduce PhysSkin, a physics-informed framework that addresses this challenge. In the spirit of Linear Blend Skinning, we learn continuous skinning fields as basis functions lifting motion subspace coordinates to full-space deformation, with subspace defined by handle transformations. To generate mesh-free, discretization-agnostic, and physically consistent skinning fields that generalize well across diverse 3D shapes, PhysSkin employs a new neural skinning fields autoencoder which consists of a transformer-based encoder and a cross-attention decoder.Furthermore, we also develop a novel physics-informed self-supervised learning strategy that incorporates on-the-fly skinning-field normalization and conflict-aware gradient correction, enabling effective balancing of energy minimization, spatial smoothness, and orthogonality constraints.PhysSkin shows outstanding performance on generalizable neural skinning and enables real-time physics-based animation.

Sixiao Zheng, Minghao Yin, Wenbo Hu, Xiaoyu Li, Ying Shan, Yanwei Fu

Video world models aim to simulate dynamic, real-world environments, yet existing methods struggle to provide unified and precise control over camera and multi-object motion, as videos inherently capture dynamics in the projected 2D image plane. To bridge this gap, we introduce VerseCrafter, a geometry-driven video world model that generates dynamic, realistic videos from a unified 4D geometric world state. Our approach is centered on a novel 4D Geometric Control representation, which encodes the world state as a static background point cloud and per-object 3D Gaussian trajectories. This representation captures each object's motion path and probabilistic 3D occupancy over time, providing a flexible, category-agnostic alternative to rigid bounding boxes and parametric models. We render 4D Geometric Control into 4D control maps for a pretrained video diffusion model, enabling high-fidelity, view-consistent video generation that faithfully follows the specified dynamics. To enable training at scale, we develop an automatic data engine and construct VerseControl4D, a real-world dataset of 35K training samples with automatically derived prompts and rendered 4D control maps. Extensive experiments show that VerseCrafter achieves superior visual quality and more accurate control over camera and multi-object motion than prior methods.

Hongyi Pan, Gorkem Durak, Halil Ertugrul Aktas, Andrea M. Bejar, Baver Tutun, Emre Uysal, Ezgi Bulbul, Mehmet Fatih Dogan, Berrin Erok, Berna Akkus Yildirim 等

Publicly available full-field digital mammography (FFDM) datasets remain limited in size, clinical annotations, and vendor diversity, hindering the development of robust models. We introduce LUMINA, a curated, multi-vendor FFDM dataset that explicitly encodes acquisition energy and vendor metadata to capture clinically relevant appearance variations often overlooked in existing benchmarks. This dataset contains 1824 images from 468 patients (960 benign, 864 malignant), with pathology-confirmed labels, BI-RADS assessments, and breast-density annotations. LUMINA spans six acquisition systems and includes both high- and low-energy imaging styles, enabling systematic analysis of vendor- and energy-induced domain shifts. To address these variations, we propose a foreground-only pixel-space alignment method ("energy harmonization") that maps images to a low-energy reference while preserving lesion morphology. We benchmark CNN and transformer models on three clinically relevant tasks: diagnosis (benign vs. malignant), BI-RADS classification, and density estimation. Two-view models consistently outperform single-view models. EfficientNet-B0 achieves an AUC of 93.54% for diagnosis, while Swin-T achieves the best macro-AUC of 89.43% for density prediction. Harmonization improves performance across architectures and produces more localized Grad-CAM responses. Overall, LUMINA provides (1) a vendor-diverse benchmark and (2) a model-agnostic harmonization framework for reliable and deployable mammography AI.

Pei Liu, Qingtian Ning, Xinyan Lu, Haipeng Liu, Weiliang Ma, Dangen She, Xianpeng Lang, Jun Ma

The pursuit of autonomous agents with predictive cognitive world models is hindered by a fundamental flaw in current vision-language models (VLMs): they lack cognitive inertia. Operating on isolated snapshots, these models cannot form a temporally coherent world view, leading to erratic decision jitter and a failure to execute complex, multi-step maneuvers. To remedy this, we introduce CogDriver, a framework designed to build a coherent world model by instilling this crucial cognitive property. Our work makes two key contributions: (1) We present CogDriver-Data, a large-scale vision-language-action dataset whose narrative annotations provide the supervisory signal for learning the temporal dynamics of a world model. (2) We develop the CogDriver-Agent, an architecture featuring a sparse temporalmemory to maintain a stable internal state, the foundation of a world model. This is enabled by a spatiotemporal knowledge distillation approach that explicitly teaches decision coherence. Comprehensive experiments validate our paradigm: CogDriver-Agent achieves a 22% increase in the closed-loop Driving Score on Bench2Drive and a 21% reduction in mean L2 error on nuScenes, establishing a new state-of-the-art. These significant gains in both long-term decision-making and imitation accuracy provide strong evidence that our agent is developing a more stable internal world model.