arXiv subjects

Xinjiang Wang

Publications and source records attributed to Xinjiang Wang.

At least 19 recordsLinked to original sources

arXiv2026

GSExplorer: Interactive World Generation with a Persistent 3D Gaussian World State

Interactive world generation requires both accurate local control and stable scene consistency over long rollouts. Existing methods improve local generation through camera or geometry conditioning and extend historical context with memory mechanisms. However, long-horizon exploration remains challenging as target-view guidance becomes incomplete and inconsistencies accumulate across repeated scene updates. We present \textbf{GSExplorer}, a generation--reconstruction framework built around a persistent 3D Gaussian world state. At each step, the current 3DGS scene provides multimodal rendering conditions for a condition-driven video generator, while camera geometry is injected through a separate Relative Ray Encoding branch. Generated videos are then fed back to a SLAM-based reconstruction backend, where inconsistency filtering prevents conflicting content from corrupting the scene and cross-chunk bundle adjustment improves spatial alignment across generation steps. We further design a data-construction pipeline that reproduces realistic 3DGS degradation during scene extrapolation, yielding 293K training pairs of degraded multimodal renderings and real target videos. GSExplorer achieves the best overall performance among the compared methods on both single-step and three-step WorldScore evaluation, with strong camera control and consistency across both settings, including a Camera Control score of 99.07 in single-step extrapolation. On 481-frame leave-and-return rollouts, it also shows strong revisit consistency, demonstrating the value of maintaining a reliable 3D world state for continued exploration.

cs.CV↗

arXiv2026

Mage-VL-Cowork: A Full Training Recipe to Learn Hybrid Computer Use From Evolved Harnesses

Effective computer-use agents (CUAs) should seamlessly combine visual GUI interaction with code and terminal execution within a single policy. We present Mage-VL-Cowork, a 4B hybrid CUA trained from a general multimodally aligned checkpoint through an end-to-end, progressive five-stage curriculum across midtraining, SFT, and RL. To power this curriculum at scale, Mage-CUA-Forge replaces per-task agent construction with reusable, skill-conditioned generation workflows and application-specific preparers, synthesizing 293,806 desktop and web tasks across 6,915 skills while reducing recurring per-task expenditure by 82.6%. Rather than directly distilling a default teacher, we evolve the teacher harness and instantiate specialized CUA harnesses to yield high-quality trajectories alongside stage-specific supervision. Mage-VL-Cowork-4B finally climbs to 66.0% on OSWorld-Verified through different training stages with more efficient reasoning cost, achieving state-of-the-art performance at its parameter size and pushing forward the frontier of on-device intelligence.

cs.CL↗

arXiv2026

Write Once, Query Many: Evolving Compact Reusable Memory for Long Videos

Long videos often exceed the context capacity of multimodal large language models (MLLMs), while agentic search repeats costly video processing and grounding for each query. Reusable memory can amortize this cost, but must preserve information for questions unknown during construction. We propose Video-EMem, an evolvable video memory harness built on time-anchored perceptual observations. This representation combines compact free-form text with sparse images to preserve semantic content and complementary visual details across temporal scales, supporting temporal navigation without a fixed semantic schema. A Writer encodes each video once by consolidating overlapping multiscale observations, and a Reader retrieves and organizes evidence solely from the resulting memory. With the underlying models frozen, we evolve their workflows and operators using distinct feedback: query-independent source fidelity and coverage for the Writer, and QA failures in evidence localization and utilization for the Reader. Experiments support this separation for generalization beyond development queries, with our evolution strategy improving average accuracy by 7.2% over the unevolved harness. While reducing storage by about 99.86% on average relative to raw videos, Video-EMem improves Qwen3-VL-30B by an average of 2.6% across three long-video QA benchmarks over direct input of 2,048 frames. Using the same o3 answer model as the state-of-the-art agentic method, it improves average accuracy by 2.5% with lower per-query latency.

cs.CV↗

arXiv2026

Beyond Semantic Organization: Memory as Execution State Management for Long-Horizon Agents

LLM-based agents increasingly tackle long-horizon tasks with interdependent decisions, where each action reshapes future constraints and intermediate errors can cascade. Existing RAG and agent memory systems organize histories by semantic similarity, retrieving content-relevant entries at decision time. We argue that this design mismatches execution-state dependencies: it fragments decision trajectories and mixes valid and erroneous traces, hindering coherent state reconstruction and error isolation. We propose MAGE (Memory as Agent-Guided Exploration), an active execution-state manager that stores interactions in a hierarchical state tree. The agent derives its state from the active root-to-current path, combining subgoal summaries, recent traces, and hints from prior branches. Four coupled operations maintain the tree: Grow records new traces, Compress summarizes completed subgoals, Maintain validates summaries, and Revise restores a target boundary and resumes on a new branch. This design bounds context growth while preserving state integrity and isolating flawed segments from the active path. Experiments on MemoryArena show that MAGE improves the average task success rate by 7.8--20.4 pp over baselines, while reducing token consumption by 55.1%. Code is available at: https://github.com/Komorebi660/MAGE.

cs.AI↗

arXiv2026

Ego4WAM: What Matters When Scaling Egocentric Human Data for Robot Learning?

Egocentric human data provides a scalable source of experience for robot learning, but varies substantially in human-robot alignment, behavioral coverage, and available supervision. Existing work shows favorable scaling with increasing human data, but it remains unclear which data properties drive downstream robot gains and how to use such data throughout the training pipeline. We present a systematic study of egocentric human data with different alignment and supervision under a unified world-action model framework. With the model backbone fixed, we disentangle the effects of human-robot alignment, data duration and task diversity, action supervision, and data usage strategies. We find that aligned human demonstrations substantially improve out-of-distribution generalization and reduce target-task robot data requirements; data duration and task diversity affect downstream capabilities differently; and video-only supervision remains effective without action labels, providing a strong foundation for subsequent video-action training. We validate these findings through closed-loop policy evaluation on both real robots and RoboDojo. Rather than treating data duration as the sole scaling axis, Ego4WAM shows how alignment, task diversity, available supervision, and usage strategy jointly shape the value of egocentric human data for robot learning.

cs.RO↗

arXiv2026

CogWAM: Aligning Semantic Cognition with World Action Modeling via Event-Driven Interfaces

Robot policies increasingly incorporate semantic reasoning and future-world prediction, yet combining these capabilities does not guarantee that local predictions and actions remain aligned with task progress. We introduce CogWAM, a cognition-guided world-action model that establishes an explicit semantic interface between task reasoning and world-action learning through a persistent Semantic State, which stores completed task events and the active subtask. CogWAM updates this state only when observations indicate semantic transitions, allowing task-level context to persist across multiple action chunks. To bridge semantic context with physical prediction and control, CogWAM employs progress-conditioned WORLD and ACTION queries that selectively extract task-relevant information for future-world prediction and action generation. During training, the Semantic State provides shared task-progress context for both branches, while inference removes the future-prediction branch and directly generates actions from observations and the maintained state. We further introduce semantic training strategies to improve transition learning and closed-loop conditioning. Without additional robot-action pretraining, CogWAM achieves 15.56 / 11.70 % Score/SR on RoboDojo and state-of-the-art performance on BiCoord, while real-world experiments demonstrate closed-loop dual-arm manipulation with 16.4 fewer Semantic State regenerations than step-wise updating.

cs.RO↗

arXiv2026

Rethinking Representations for World-Action Modeling

World-action models jointly learn robot policies and predict future observations, making the representation space an interface between control and prediction. We study the design of this space through controlled comparisons, finding that neither reconstruction fidelity nor pre-trained perceptual features alone ensure effective policy learning. These findings motivate ReWAM, a representation-centric world-action model built on pre-trained DINO features. Feature Calibration and a Temporal Representation Bottleneck organize these features into compact world states suited to dynamics modeling. Action-Grounded Representation Shaping routes only action-loss gradients to the bottleneck, thereby letting the policy shape what the representation encodes while the world model learns how it evolves. Without generative video pre-training, ReWAM achieves 93.6% success on RoboTwin 2.0. On RoboDojo, it achieves an average score of 12.29 and a success rate of 8.28% using approximately 600 hours of embodied pre-training data.

cs.CV↗

arXiv2026

MeRoPE: Metric Rotary Position Embedding for Camera-Controlled Video Generation

In camera-controlled video generation, geometry-aware positional encodings condition tokens on camera extrinsics and per-token viewing rays. Existing schemes, however, have a scale-dependent failure mode on real-world metric camera trajectories: homogeneous projective encodings cause attention logits and feature norms to grow unbounded with physical translation baselines. We propose MeRoPE (Metric Rotary Position Embedding), a norm-preserving relative camera encoding for attention. MeRoPE encodes relative orientations between calibrated viewing rays with orthogonal rotation blocks, maps raw metric displacements into multi-frequency rotary phases, and adds a disparity-anchored correspondence prior along the epipolar arc. This design strictly preserves feature norms, bounds pre-softmax attention logits regardless of the physical translation scale, and maintains exact invariance to global rigid coordinate changes. Across nuScenes and PanShot, which cover large-baseline trajectories and diverse camera optics, respectively, MeRoPE achieves stronger camera control than prior encodings, with the best consistency between generated camera motion and conditioning poses in both rotation and translation. Code will be made publicly available.

cs.CV↗

arXiv2026

MESA:Task-Adaptive Multi-Structure Evidence Selection for Long-Horizon Agent Memory

Long-horizon agents accumulate trajectories spanning hundreds of interleaved reasoning, action, and observation steps, where answering a query may depend on evidence buried far back in the history. External memory stores such trajectories as structured representations, yet each structure provides a distinct and incomplete view. Existing multi-memory systems either read a fixed set of structures for every query, inflating context and introducing noise, or route each query to a single structure, preventing the composition of complementary evidence. A controlled analysis on AMA-Bench shows that the optimal memory configuration is typically neither a single structure nor the full union, but a tailored composition of multiple structural memories that varies with query and task demands. Motivated by these findings, we formulate structure-level dynamic selection: selecting and fusing a query-adaptive subset from a library of specialized memory structures. We propose MESA (a Multi-structure Evidence Selection framework for long-horizon Agent), which builds five complementary structure views of each trajectory and learns from end-to-end answer-level feedback to select and fuse a query-specific subset for a frozen answer model. To learn under this weak supervision, MESA employs harness optimization with prior-guided search and UCB-guided scheduling to balance exploration and exploitation. On AMA-Bench, MESA outperforms the strongest baseline by 8.5% while using 41% fewer evidence tokens than the all-structure alternative.

cs.AI↗

arXiv2026

SkillHEX: Improving Agent Skills via Hypothesis-Driven Autonomous Exploration and Exploitation

Although agent skills equip LLMs with reusable procedural knowledge, manual maintenance suffers from high costs, unscalability, and misalignment. Real-world deployments thus require autonomous, on-demand skill evolution at test time, constrained by limited interaction budgets and a lack of training or validation sets. This setting introduces a severe sparse reward challenge, where outcomes conflate multiple latent failure causes. Under such ambiguity, existing methods that greedily refine a single incumbent skill are particularly vulnerable to an exploitation trap, allowing early misdiagnoses to exhaust limited trials along unproductive trajectories. To address this, we introduce SkillHEX, a closed-loop framework coupling hypothesis-driven self-verification with evidence-guided tree search. SkillHEX translates falsifiable failure hypotheses into executable tests, producing diagnostic evidence as dense reward without additional environment attempts. This evidence guides a search over persistent skill-revision branches, dynamically balancing the exploitation of supported edits with the exploration of plausible alternatives. Evaluated on 87 tasks from SkillsBench, SkillHEX outperforms existing self-evolving methods and achieves an average pass rate of 55.9% and 57.9% using GPT-5.3-Codex and Claude Opus 4.7 under a five-iteration budget, respectively.

cs.AI↗

arXiv2026

AeroAct: Action-Centered World-Action Models for Language-Conditioned Quadrotor Flight

Language-conditioned quadrotor flight requires a policy to ground semantic goals, anticipate the visual consequences of ego-motion, and output control references that remain smooth and dynamically executable under rapidly changing first-person views. Existing aerial vision-language navigation and vision-language-action methods commonly use discrete actions, high-level waypoints, or instantaneous velocity commands, which provide limited supervision about how flight actions change future observations. We present AeroAct, an action-centered world-action model (WAM) for quadrotor navigation. To the best of our knowledge, AeroAct is the first WAM instantiated and demonstrated for real-world aerial flight. The model adapts a pretrained video diffusion Transformer to predict local trajectory-action chunks from egocentric visual history, proprioception, and language. Future first-person frames are used during training as dense consequence supervision, while deployment directly decodes actions without generating future video. To obtain aligned visual, state, language, and dynamically feasible action data, we build a DiffAero-based pipeline with complementary Isaac Lab and 3D Gaussian splatting renderers. We further introduce a low-cost handheld collection device that couples camera observations with motion estimates to recreate flight-like egocentric trajectories, and a self-guidance procedure that improves temporal consistency across overlapping trajectory chunks. Closed-loop simulation and real-world experiments show that temporal visual context improves target tracking and object-search performance, and that WAM-based policies can be executed on a physical quadrotor.

cs.RO↗

arXiv2026

TriggerBench: Investigating Prospective Memory for Large Language Models

While Large Language Models (LLMs) are increasingly deployed in long interactions, existing evaluations focus predominantly on retrospective memory (RM) via explicit queries. Prospective memory (PM), the critical ability to spontaneously recall and act on latent constraints without direct prompts, remains largely unevaluated. We introduce TriggerBench, a comprehensive PM benchmark spanning five dimensions across both daily assistants and professional workflows. TriggerBench pairs scenarios with matched RM controls, contrastive positive/negative variants, and overloaded triggers, enabling fine-grained measurement of proactive recall, false-alarm rate, and attentional robustness under a single protocol. Our evaluation yields three key findings. (i) PM shows a precision-recall trade-off and attentional fragility. Though enhanced reasoning significantly improves proactive recall, models may overfit to an "always-remind" heuristic. Furthermore, PM accuracy degrades substantially under implicit constraints or triggers overloaded by concurrent user requests, indicating that robust PM remains an open challenge. (ii) PM is notably harder than RM: on identical contexts, RM near-saturates up to 100K tokens, while PM decays sharply as context length scales. (iii) PM may serve as a behavioral probe of spare reasoning capacity. Pairing PM scenarios with AIME-2025 math problems reveals that successful trajectories yield higher PM accuracy than failed ones at the same context length, showing PM tracks spare reasoning budget that token count obscures. Project page: https://github.com/KristenZHANG/TriggerBench-Official.

cs.CL↗

arXiv2025

Text4Seg: Reimagining Image Segmentation as Text Generation

Multimodal Large Language Models (MLLMs) have shown exceptional capabilities in vision-language tasks; however, effectively integrating image segmentation into these models remains a significant challenge. In this paper, we introduce Text4Seg, a novel text-as-mask paradigm that casts image segmentation as a text generation problem, eliminating the need for additional decoders and significantly simplifying the segmentation process. Our key innovation is semantic descriptors, a new textual representation of segmentation masks where each image patch is mapped to its corresponding text label. This unified representation allows seamless integration into the auto-regressive training pipeline of MLLMs for easier optimization. We demonstrate that representing an image with $16\times16$ semantic descriptors yields competitive segmentation performance. To enhance efficiency, we introduce the Row-wise Run-Length Encoding (R-RLE), which compresses redundant text sequences, reducing the length of semantic descriptors by 74% and accelerating inference by $3\times$, without compromising performance. Extensive experiments across various vision tasks, such as referring expression segmentation and comprehension, show that Text4Seg achieves state-of-the-art performance on multiple datasets by fine-tuning different MLLM backbones. Our approach provides an efficient, scalable solution for vision-centric tasks within the MLLM framework.

cs.CV↗

arXiv2024

Revisiting the Integration of Convolution and Attention for Vision Backbone

Convolutions (Convs) and multi-head self-attentions (MHSAs) are typically considered alternatives to each other for building vision backbones. Although some works try to integrate both, they apply the two operators simultaneously at the finest pixel granularity. With Convs responsible for per-pixel feature extraction already, the question is whether we still need to include the heavy MHSAs at such a fine-grained level. In fact, this is the root cause of the scalability issue w.r.t. the input resolution for vision transformers. To address this important problem, we propose in this work to use MSHAs and Convs in parallel \textbf{at different granularity levels} instead. Specifically, in each layer, we use two different ways to represent an image: a fine-grained regular grid and a coarse-grained set of semantic slots. We apply different operations to these two representations: Convs to the grid for local features, and MHSAs to the slots for global features. A pair of fully differentiable soft clustering and dispatching modules is introduced to bridge the grid and set representations, thus enabling local-global fusion. Through extensive experiments on various vision tasks, we empirically verify the potential of the proposed integration scheme, named \textit{GLMix}: by offloading the burden of fine-grained features to light-weight Convs, it is sufficient to use MHSAs in a few (e.g., 64) semantic slots to match the performance of recent state-of-the-art backbones, while being more efficient. Our visualization results also demonstrate that the soft clustering module produces a meaningful semantic grouping effect with only IN1k classification supervision, which may induce better interpretability and inspire new weakly-supervised semantic segmentation approaches. Code will be available at \url{https://github.com/rayleizhu/GLMix}.

cs.CV↗

arXiv2024

ProxyCLIP: Proxy Attention Improves CLIP for Open-Vocabulary Segmentation

Open-vocabulary semantic segmentation requires models to effectively integrate visual representations with open-vocabulary semantic labels. While Contrastive Language-Image Pre-training (CLIP) models shine in recognizing visual concepts from text, they often struggle with segment coherence due to their limited localization ability. In contrast, Vision Foundation Models (VFMs) excel at acquiring spatially consistent local visual representations, yet they fall short in semantic understanding. This paper introduces ProxyCLIP, an innovative framework designed to harmonize the strengths of both CLIP and VFMs, facilitating enhanced open-vocabulary semantic segmentation. ProxyCLIP leverages the spatial feature correspondence from VFMs as a form of proxy attention to augment CLIP, thereby inheriting the VFMs' robust local consistency and maintaining CLIP's exceptional zero-shot transfer capacity. We propose an adaptive normalization and masking strategy to get the proxy attention from VFMs, allowing for adaptation across different VFMs. Remarkably, as a training-free approach, ProxyCLIP significantly improves the average mean Intersection over Union (mIoU) across eight benchmarks from 40.3 to 44.4, showcasing its exceptional efficacy in bridging the gap between spatial precision and semantic richness for the open-vocabulary segmentation task.

cs.CV↗

arXiv2024

ClearCLIP: Decomposing CLIP Representations for Dense Vision-Language Inference

Despite the success of large-scale pretrained Vision-Language Models (VLMs) especially CLIP in various open-vocabulary tasks, their application to semantic segmentation remains challenging, producing noisy segmentation maps with mis-segmented regions. In this paper, we carefully re-investigate the architecture of CLIP, and identify residual connections as the primary source of noise that degrades segmentation quality. With a comparative analysis of statistical properties in the residual connection and the attention output across different pretrained models, we discover that CLIP's image-text contrastive training paradigm emphasizes global features at the expense of local discriminability, leading to noisy segmentation results. In response, we propose ClearCLIP, a novel approach that decomposes CLIP's representations to enhance open-vocabulary semantic segmentation. We introduce three simple modifications to the final layer: removing the residual connection, implementing the self-self attention, and discarding the feed-forward network. ClearCLIP consistently generates clearer and more accurate segmentation maps and outperforms existing approaches across multiple benchmarks, affirming the significance of our discoveries.

cs.CV↗

arXiv2024

RelayAttention for Efficient Large Language Model Serving with Long System Prompts

A practical large language model (LLM) service may involve a long system prompt, which specifies the instructions, examples, and knowledge documents of the task and is reused across requests. However, the long system prompt causes throughput/latency bottlenecks as the cost of generating the next token grows w.r.t. the sequence length. This paper aims to improve the efficiency of LLM services that involve long system prompts. Our key observation is that handling these system prompts requires heavily redundant memory accesses in existing causal attention computation algorithms. Specifically, for batched requests, the cached hidden states (\ie, key-value pairs) of system prompts are transferred from off-chip DRAM to on-chip SRAM multiple times, each corresponding to an individual request. To eliminate such a redundancy, we propose RelayAttention, an attention algorithm that allows reading these hidden states from DRAM exactly once for a batch of input tokens. RelayAttention is a free lunch: it maintains the generation quality while requiring no model retraining, as it is based on a mathematical reformulation of causal attention. We have observed significant performance improvements to a production-level system, vLLM, through integration with RelayAttention. The improvements are even more profound with longer system prompts.

cs.CL↗

arXiv2024

An Open and Comprehensive Pipeline for Unified Object Grounding and Detection

Grounding-DINO is a state-of-the-art open-set detection model that tackles multiple vision tasks including Open-Vocabulary Detection (OVD), Phrase Grounding (PG), and Referring Expression Comprehension (REC). Its effectiveness has led to its widespread adoption as a mainstream architecture for various downstream applications. However, despite its significance, the original Grounding-DINO model lacks comprehensive public technical details due to the unavailability of its training code. To bridge this gap, we present MM-Grounding-DINO, an open-source, comprehensive, and user-friendly baseline, which is built with the MMDetection toolbox. It adopts abundant vision datasets for pre-training and various detection and grounding datasets for fine-tuning. We give a comprehensive analysis of each reported result and detailed settings for reproduction. The extensive experiments on the benchmarks mentioned demonstrate that our MM-Grounding-DINO-Tiny outperforms the Grounding-DINO-Tiny baseline. We release all our models to the research community. Codes and trained models are released at https://github.com/open-mmlab/mmdetection/tree/main/configs/mm_grounding_dino.

cs.CV↗