arXiv subjects

Wenjun Li

Publications and source records attributed to Wenjun Li.

At least 19 recordsLinked to original sources

arXiv2026

Towards Precise Intent-Aligned VLA Aerial Navigation via Expert-Guided GRPO

Vision-Language-Action (VLA) models offer a promising end-to-end paradigm for unmanned aerial vehicles (UAVs) to accomplish complex tasks specified by fine-grained instructions. However, standard supervised fine-tuning (SFT) suffers from data scarcity, limited generalization, and weak supervision for nuanced and complicated human intents. Reinforcement fine-tuning offers a natural way to mitigate these challenges and align policy behaviors with human intents through designable feedback, but applying it to aerial navigation remains challenging due to inefficient exploration in expansive continuous spaces. To address these challenges, we introduce an efficient reinforcement learning (RL) framework for VLA-based aerial navigation. At its core, we propose EG-GRPO (Expert-Guided Group Relative Policy Optimization) to augment online rollouts with few-shot expert data. Additionally, we design a heterogeneous pipeline enabling parallel simulation and inference, which reduces rollout time by 43.5%. Across multiple tasks specified by complex human intents, EG-GRPO improves the success rate to 2.13x that of the SFT baseline, while improving intent alignment performance by 60.9%. These results demonstrate that our framework can move aerial navigation toward precise intent-aligned flight.

cs.RO↗

arXiv2026

HiSem: Hierarchical Semantic Disentangling for Remote Sensing Image Change Captioning

Remote sensing image change captioning (RSICC) aims to achieve high-level semantic understanding of genuine changes occurring between bi-temporal images. Despite notable progress, existing methods are fundamentally limited by a shared modeling assumption: changed and unchanged image pairs, which have intrinsically different semantic granularities, are processed under a unified modeling strategy. This modeling inconsistency leads to semantic entanglement between coarse-grained change existence judgment and fine-grained semantic understanding.To address the above limitation, we propose a novel hierarchical semantic disentangling network (HiSem) that explicitly disentangles semantic representations of different granularities. Specifically, we first introduce the Bidirectional Differential Attention Modulation (BDAM) module that leverages discrepancy-aware attention to enhance cross-temporal interactions, thereby amplifying true change signals while suppressing irrelevant variations. Building upon this, we design a Hierarchical Adaptive Semantic Disentanglement (HASD) module that performs adaptive routing at two hierarchical levels: a coarse-grained image-level routing mechanism distinguishes changed and unchanged image pairs, while a fine-grained token-level Mixture-of-Experts (MoE) block models diverse and heterogeneous change semantics for changed samples. Extensive experiments on two benchmark datasets demonstrate that HiSem outperfoms previous methods, achieving a significant improvement of +7.52\% BLEU-4 on the WHU-CDC dataset. More importantly, our approach provides a structured perspective for RSICC by explicitly aligning model design with the intrinsic semantic heterogeneity of bi-temporal scenes. The code will be available at https://github.com/Man-Wang-star/HiSem

cs.CV↗

arXiv2026

LeHome: A Simulation Environment for Deformable Object Manipulation in Household Scenarios

Household environments present one of the most common, impactful yet challenging application domains for robotics. Within household scenarios, manipulating deformable objects is particularly difficult, both in simulation and real-world execution, due to varied categories and shapes, complex dynamics, and diverse material properties, as well as the lack of reliable deformable-object support in existing simulations. We introduce LeHome, a comprehensive simulation environment designed for deformable object manipulation in household scenarios. LeHome covers a wide spectrum of deformable objects, such as garments and food items, offering high-fidelity dynamics and realistic interactions that existing simulators struggle to simulate accurately. Moreover, LeHome supports multiple robotic embodiments and emphasizes low-cost robots as a core focus, enabling end-to-end evaluation of household tasks on resource-constrained hardware. By bridging the gap between realistic deformable object simulation and practical robotic platforms, LeHome provides a scalable testbed for advancing household robotics. Webpage: https://lehome-web.github.io/ .

cs.RO↗

arXiv2026

HiMAP-Travel: Hierarchical Multi-Agent Planning for Long-Horizon Constrained Travel

Sequential LLM agents fail on long-horizon planning with hard constraints like budgets and diversity requirements. As planning progresses and context grows, these agents drift from global constraints. We propose HiMAP-Travel, a hierarchical multi-agent framework that splits planning into strategic coordination and parallel day-level execution. A Coordinator allocates resources across days, while Day Executors plan independently in parallel. Three key mechanisms enable this: a transactional monitor enforcing budget and uniqueness constraints across parallel agents, a bargaining protocol allowing agents to reject infeasible sub-goals and trigger re-planning, and a single policy trained with GRPO that powers all agents through role conditioning. On TravelPlanner, HiMAP-Travel with Qwen3-8B achieves 52.78% validation and 52.65% test Final Pass Rate (FPR). In a controlled comparison with identical model, training, and tools, it outperforms the sequential DeepTravel baseline by +8.67~pp. It also surpasses ATLAS by +17.65~pp and MTP by +10.0~pp. On FlexTravelBench multi-turn scenarios, it achieves 44.34% (2-turn) and 37.42% (3-turn) FPR while reducing latency 2.5x through parallelization.

cs.AI↗

arXiv2026

Discern Truth from Falsehood: Reducing Over-Refusal via Contrastive Refinement

Large language models (LLMs) aligned for safety often suffer from over-refusal, the tendency to reject seemingly toxic or benign prompts by misclassifying them as toxic. This behavior undermines models' helpfulness and restricts usability in sensitive or nuanced contexts. While prior work has proposed mitigation strategies such as data augmentation and activation steering, these approaches often face a trade-off: reducing over-refusal typically degrades the model's ability to reject genuinely harmful content. We argue that this issue arises from the ambiguous influence of toxic and seemingly toxic prompts on the model's learning dynamics. To address it, we introduce a preceding alignment stage, DCR: Discernment via Contrastive Refinement. Both theoretically and empirically, we demonstrate that contrastive refinement improves an LLM's capacity to distinguish truly toxic prompts from superficially toxic ones. Evaluation across diverse benchmarks shows that our method effectively reduces over-refusal while preserving the safety benefits of alignment. Importantly, it achieves this with minimal degradation of general capabilities, offering a more principled and robust direction for safety alignment.

cs.CL↗

arXiv2026

SRR-Judge: Step-Level Rating and Refinement for Enhancing Search-Integrated Reasoning in Search Agents

Recent deep search agents built on large reasoning models (LRMs) excel at complex question answering by iteratively planning, acting, and gathering evidence, a capability known as search-integrated reasoning. However, mainstream approaches often train this ability using only outcome-based supervision, neglecting the quality of intermediate thoughts and actions. We introduce SRR-Judge, a framework for reliable step-level assessment of reasoning and search actions. Integrated into a modified ReAct-style rate-and-refine workflow, SRR-Judge provides fine-grained guidance for search-integrated reasoning and enables efficient post-training annotation. Using SRR-annotated data, we apply an iterative rejection sampling fine-tuning procedure to enhance the deep search capability of the base agent. Empirically, SRR-Judge delivers more reliable step-level evaluations than much larger models such as DeepSeek-V3.1, with its ratings showing strong correlation with final answer correctness. Moreover, aligning the policy with SRR-Judge annotated trajectories leads to substantial performance gains, yielding over a 10 percent average absolute pass@1 improvement across challenging deep search benchmarks.

cs.CL↗

arXiv2025

VLA-AN: An Efficient and Onboard Vision-Language-Action Framework for Aerial Navigation in Complex Environments

This paper proposes VLA-AN, an efficient and onboard Vision-Language-Action (VLA) framework dedicated to autonomous drone navigation in complex environments. VLA-AN addresses four major limitations of existing large aerial navigation models: the data domain gap, insufficient temporal navigation with reasoning, safety issues with generative action policies, and onboard deployment constraints. First, we construct a high-fidelity dataset utilizing 3D Gaussian Splatting (3D-GS) to effectively bridge the domain gap. Second, we introduce a progressive three-stage training framework that sequentially reinforces scene comprehension, core flight skills, and complex navigation capabilities. Third, we design a lightweight, real-time action module coupled with geometric safety correction. This module ensures fast, collision-free, and stable command generation, mitigating the safety risks inherent in stochastic generative policies. Finally, through deep optimization of the onboard deployment pipeline, VLA-AN achieves a robust real-time 8.3x improvement in inference throughput on resource-constrained UAVs. Extensive experiments demonstrate that VLA-AN significantly improves spatial grounding, scene reasoning, and long-horizon navigation, achieving a maximum single-task success rate of 98.1%, and providing an efficient, practical solution for realizing full-chain closed-loop autonomy in lightweight aerial robots.

cs.RO↗

arXiv2025

Doc-Researcher: A Unified System for Multimodal Document Parsing and Deep Research

Deep Research systems have revolutionized how LLMs solve complex questions through iterative reasoning and evidence gathering. However, current systems remain fundamentally constrained to textual web data, overlooking the vast knowledge embedded in multimodal documents Processing such documents demands sophisticated parsing to preserve visual semantics (figures, tables, charts, and equations), intelligent chunking to maintain structural coherence, and adaptive retrieval across modalities, which are capabilities absent in existing systems. In response, we present Doc-Researcher, a unified system that bridges this gap through three integrated components: (i) deep multimodal parsing that preserves layout structure and visual semantics while creating multi-granular representations from chunk to document level, (ii) systematic retrieval architecture supporting text-only, vision-only, and hybrid paradigms with dynamic granularity selection, and (iii) iterative multi-agent workflows that decompose complex queries, progressively accumulate evidence, and synthesize comprehensive answers across documents and modalities. To enable rigorous evaluation, we introduce M4DocBench, the first benchmark for Multi-modal, Multi-hop, Multi-document, and Multi-turn deep research. Featuring 158 expert-annotated questions with complete evidence chains across 304 documents, M4DocBench tests capabilities that existing benchmarks cannot assess. Experiments demonstrate that Doc-Researcher achieves 50.6% accuracy, 3.4xbetter than state-of-the-art baselines, validating that effective document research requires not just better retrieval, but fundamentally deep parsing that preserve multimodal integrity and support iterative research. Our work establishes a new paradigm for conducting deep research on multimodal document collections.

cs.IR↗

arXiv2025

Embodied Navigation Foundation Model

Navigation is a fundamental capability in embodied AI, representing the intelligence required to perceive and interact within physical environments following language instructions. Despite significant progress in large Vision-Language Models (VLMs), which exhibit remarkable zero-shot performance on general vision-language tasks, their generalization ability in embodied navigation remains largely confined to narrow task settings and embodiment-specific architectures. In this work, we introduce a cross-embodiment and cross-task Navigation Foundation Model (NavFoM), trained on eight million navigation samples that encompass quadrupeds, drones, wheeled robots, and vehicles, and spanning diverse tasks such as vision-and-language navigation, object searching, target tracking, and autonomous driving. NavFoM employs a unified architecture that processes multimodal navigation inputs from varying camera configurations and navigation horizons. To accommodate diverse camera setups and temporal horizons, NavFoM incorporates identifier tokens that embed camera view information of embodiments and the temporal context of tasks. Furthermore, to meet the demands of real-world deployment, NavFoM controls all observation tokens using a dynamically adjusted sampling strategy under a limited token length budget. Extensive evaluations on public benchmarks demonstrate that our model achieves state-of-the-art or highly competitive performance across multiple navigation tasks and embodiments without requiring task-specific fine-tuning. Additional real-world experiments further confirm the strong generalization capability and practical applicability of our approach.

cs.RO↗

arXiv2025

Reinforcement Learning Foundations for Deep Research Systems: A Survey

Deep research systems, agentic AI that solve complex, multi-step tasks by coordinating reasoning, search across the open web and user files, and tool use, are moving toward hierarchical deployments with a Planner, Coordinator, and Executors. In practice, training entire stacks end-to-end remains impractical, so most work trains a single planner connected to core tools such as search, browsing, and code. While SFT imparts protocol fidelity, it suffers from imitation and exposure biases and underuses environment feedback. Preference alignment methods such as DPO are schema and proxy-dependent, off-policy, and weak for long-horizon credit assignment and multi-objective trade-offs. A further limitation of SFT and DPO is their reliance on human defined decision points and subskills through schema design and labeled comparisons. Reinforcement learning aligns with closed-loop, tool-interaction research by optimizing trajectory-level policies, enabling exploration, recovery behaviors, and principled credit assignment, and it reduces dependence on such human priors and rater biases. This survey is, to our knowledge, the first dedicated to the RL foundations of deep research systems. It systematizes recent work along three axes: (i) data synthesis and curation; (ii) RL methods for agentic research covering stability, sample efficiency, long context handling, reward and credit design, multi-objective optimization, and multimodal integration; and (iii) agentic RL training systems and frameworks. We also cover agent architecture and coordination, as well as evaluation and benchmarks, including recent QA, VQA, long-form synthesis, and domain-grounded, tool-interaction tasks. We distill recurring patterns, surface infrastructure bottlenecks, and offer practical guidance for training robust, transparent deep research agents with RL.

cs.AI↗

arXiv2025

Online Learning Based Efficient Resource Allocation for LoRaWAN Network

The deployment of large-scale LoRaWAN networks requires jointly optimizing conflicting metrics like Packet Delivery Ratio (PDR) and Energy Efficiency (EE) by dynamically allocating transmission parameters, including Carrier Frequency, Spreading Factor, and Transmission Power. Existing methods often oversimplify this challenge, focusing on a single metric or lacking the adaptability needed for dynamic channel environments, leading to suboptimal performance. To address this, we propose two online learning-based resource allocation frameworks that intelligently navigate the PDR-EE trade-off. Our foundational proposal, D-LoRa, is a fully distributed framework that models the problem as a Combinatorial Multi-Armed Bandit. By decomposing the joint parameter selection and employing specialized, disaggregated reward functions, D-LoRa dramatically reduces learning complexity and enables nodes to autonomously adapt to network dynamics. To further enhance performance in LoRaWAN networks, we introduce CD-LoRa, a hybrid framework that integrates a lightweight, centralized initialization phase to perform a one-time, quasi-optimal channel assignment and action space pruning, thereby accelerating subsequent distributed learning. Extensive simulations and real-world field experiments demonstrate the superiority of our frameworks, showing that D-LoRa excels in non-stationary environments while CD-LoRa achieves the fastest convergence in stationary conditions. In physical deployments, our methods outperform state-of-the-art baselines, improving PDR by up to 10.8% and EE by 26.1%, confirming their practical effectiveness for scalable and efficient LoRaWAN networks.

cs.NI↗

arXiv2025

Branched $\alpha$-combinatorial Ricci flows on closed surfaces with Euler characteristic $\chi\le 0$

In this paper we introduce the branched $\alpha$-flows on closed surfaces with Euler characteristic $\chi \leq 0$. Based on the strict convexity of the branched $\alpha$-potentials, we establish the long time existence and convergence of the solutions to the branched $\alpha$-flows, which generalizes Ge and Xu's main results \cite{2015,2015A} on the $\alpha$-flows. In addtion, we study the prescribed curvature problems under the relaxed precondition $\chi(M)\in \mathbb{Z}$ via alternative $\alpha$-flows, establishing admissibility conditions for prescribed curvatures and their exponential convergence to target metrics.

math.DG↗

arXiv2025

PSSketch: Finding Persistent and Sparse Flow with High Accuracy and Efficiency

Finding persistent sparse (PS) flow is critical to early warning of many threats. Previous works have predominantly focused on either heavy or persistent flows, with limited attention given to PS flows. Although some recent studies pay attention to PS flows, they struggle to establish an objective criterion due to insufficient data-driven observations, resulting in reduced accuracy. In this paper, we define a new criterion "anomaly boundary" to distinguish PS flows from regular flows. Specifically, a flow whose persistence exceeds a threshold will be protected, while a protected flow with a density lower than a threshold is reported as a PS flow. We then introduce PSSketch, a high-precision layered sketch to find PS flows. PSSketch employs variable-length bitwise counters, where the first layer tracks the frequency and persistence of all flows, and the second layer protects potential PS flows and records overflow counts from the first layer. Some optimizations have also been implemented to reduce memory consumption further and improve accuracy. The experiments show that PSSketch reduces memory consumption by an order of magnitude compared to the strawman solution combined with existing work. Compared with SOTA solutions for finding PS flows, it outperforms up to 2.94x in F1 score and reduces ARE by 1-2 orders of magnitude. Meanwhile, PSSketch achieves a higher throughput than these solutions.

cs.DS↗

arXiv2025

Understanding R1-Zero-Like Training: A Critical Perspective

DeepSeek-R1-Zero has shown that reinforcement learning (RL) at scale can directly enhance the reasoning capabilities of LLMs without supervised fine-tuning. In this work, we critically examine R1-Zero-like training by analyzing its two core components: base models and RL. We investigate a wide range of base models, including DeepSeek-V3-Base, to understand how pretraining characteristics influence RL performance. Our analysis reveals that DeepSeek-V3-Base already exhibit ''Aha moment'', while Qwen2.5 base models demonstrate strong reasoning capabilities even without prompt templates, suggesting potential pretraining biases. Additionally, we identify an optimization bias in Group Relative Policy Optimization (GRPO), which artificially increases response length (especially for incorrect outputs) during training. To address this, we introduce Dr. GRPO, an unbiased optimization method that improves token efficiency while maintaining reasoning performance. Leveraging these insights, we present a minimalist R1-Zero recipe that achieves 43.3% accuracy on AIME 2024 with a 7B base model, establishing a new state-of-the-art. Our code is available at https://github.com/sail-sg/understand-r1-zero.

cs.LG↗

arXiv2025

Open-Sora 2.0: Training a Commercial-Level Video Generation Model in $200k

Video generation models have achieved remarkable progress in the past year. The quality of AI video continues to improve, but at the cost of larger model size, increased data quantity, and greater demand for training compute. In this report, we present Open-Sora 2.0, a commercial-level video generation model trained for only $200k. With this model, we demonstrate that the cost of training a top-performing video generation model is highly controllable. We detail all techniques that contribute to this efficiency breakthrough, including data curation, model architecture, training strategy, and system optimization. According to human evaluation results and VBench scores, Open-Sora 2.0 is comparable to global leading video generation models including the open-source HunyuanVideo and the closed-source Runway Gen-3 Alpha. By making Open-Sora 2.0 fully open-source, we aim to democratize access to advanced video generation technology, fostering broader innovation and creativity in content creation. All resources are publicly available at: https://github.com/hpcaitech/Open-Sora.

cs.GR↗

arXiv2025

Adaptive Tool Use in Large Language Models with Meta-Cognition Trigger

Large language models (LLMs) have shown remarkable emergent capabilities, transforming the execution of functional tasks by leveraging external tools for complex problems that require specialized processing or up-to-date data. While existing research expands LLMs access to diverse tools (e.g., program interpreters, search engines, calculators), the necessity of using these tools is often overlooked, leading to indiscriminate tool invocation. This naive approach raises two key issues: increased latency due to unnecessary tool calls, and potential errors resulting from faulty interactions with external tools. In this paper, we introduce meta-cognition as a proxy for LLMs self-assessment of their capabilities, reflecting the model's awareness of its own limitations. Based on this, we propose MeCo, an adaptive decision-making strategy for external tool use. MeCo quantifies metacognitive scores by capturing high-level cognitive signals in the representation space, guiding when to invoke tools. Notably, MeCo is fine-tuning-free and incurs minimal cost. Experiments across multiple backbone models and benchmarks show that MeCo reliably detects LLMs' internal cognitive signals and significantly improves tool-use decision-making.

cs.CL↗

arXiv2025

Improving Environment Novelty Quantification for Effective Unsupervised Environment Design

Unsupervised Environment Design (UED) formalizes the problem of autocurricula through interactive training between a teacher agent and a student agent. The teacher generates new training environments with high learning potential, curating an adaptive curriculum that strengthens the student's ability to handle unseen scenarios. Existing UED methods mainly rely on regret, a metric that measures the difference between the agent's optimal and actual performance, to guide curriculum design. Regret-driven methods generate curricula that progressively increase environment complexity for the student but overlook environment novelty -- a critical element for enhancing an agent's generalizability. Measuring environment novelty is especially challenging due to the underspecified nature of environment parameters in UED, and existing approaches face significant limitations. To address this, this paper introduces the Coverage-based Evaluation of Novelty In Environment (CENIE) framework. CENIE proposes a scalable, domain-agnostic, and curriculum-aware approach to quantifying environment novelty by leveraging the student's state-action space coverage from previous curriculum experiences. We then propose an implementation of CENIE that models this coverage and measures environment novelty using Gaussian Mixture Models. By integrating both regret and novelty as complementary objectives for curriculum design, CENIE facilitates effective exploration across the state-action space while progressively increasing curriculum complexity. Empirical evaluations demonstrate that augmenting existing regret-based UED algorithms with CENIE achieves state-of-the-art performance across multiple benchmarks, underscoring the effectiveness of novelty-driven autocurricula for robust generalization.

cs.LG↗

arXiv2025

Digital-Twin assisted Network Energy Optimization during Low Traffic Hours

As wireless network technology advances towards the sixth generation (6G), increasing network energy consumption has become a critical concern due to the growing demand for diverse services, radio deployments at various frequencies, larger bandwidths, and more antennas. Network operators must manage energy usage not only to reduce operational cost and improve revenue but also to minimize environmental impact by reducing the carbon footprint. The 3rd Generation Partnership Project (3GPP) has introduced several network energy savings (NES) features. However, the implementation details and system-level aspects of these features have not been thoroughly investigated. In this paper, we explore system-level resource optimization for network energy savings in low-traffic scenarios. We introduce multiple NES optimization formulations and strategies, and further analyze their performance using a detailed network digital twin. Our results demonstrate promising NES gains of up to 44%. Additionally, we provide practical considerations for implementing the proposed schemes and examine their impacts on user equipment (UE) operation.

cs.NI↗