Deguo Xia

Deguo Xia, Zihan Li, Haochen Zhao et al.

Lane-level maps are critical infrastructure for autonomous driving and lane-level navigation, yet constructing and maintaining standardized lane networks for hundreds of cities remains highly labor-intensive. Recent end-to-end vectorized mapping methods can predict lane geometry and topology directly from sensor data, but they typically treat mapping specifications and traffic regulations as implicit, dataset-dependent supervision. Moreover, in complex scenes (e.g., worn or missing markings and occlusions), correct lane configurations are often under-determined by visual evidence alone, making specification violations a major source of human post-editing. We propose MapAgent, an industrial-grade agentic architecture that augments a vectorization backbone for specification-compliant lane-map production. Rather than merely adding an agent loop to map prediction, MapAgent couples backbone perception with explicit specification verification, constraint-aware reasoning, and deterministic map editing under a bounded, verification-driven Judge-Planner-Worker loop. A vision-language Judge diagnoses errors by jointly inspecting visual evidence and draft vectors, while a tool-calling Planner generates minimal corrective edits with post-edit re-validation. To remain scalable for city-scale production, MapAgent is selectively triggered only on tiles with low backbone confidence, adding modest overhead while preserving throughput. Experiments on real-world datasets show consistent gains over strong production baselines, especially in complex and long-tail scenarios. Additionally, MapAgent has been integrated into Baidu Maps, supporting lane-level map generation for over 360 cities nationwide and elevating the overall production automation to over 95%, demonstrating MapAgent's practicality and effectiveness for large-scale lane-level map generation.

Rui Zhu, Xin Shen, Shuchen Wu et al.

Spatial reasoning has emerged as a critical capability for Multimodal Large Language Models (MLLMs), drawing increasing attention and rapid advancement. However, existing benchmarks primarily focus on single-step perception-to-judgment tasks, leaving scenarios requiring complex visual-spatial logical chains significantly underexplored. To bridge this gap, we introduce Video-MSR, the first benchmark specifically designed to evaluate Multi-hop Spatial Reasoning (MSR) in dynamic video scenarios. Video-MSR systematically probes MSR capabilities through four distinct tasks: Constrained Localization, Chain-based Reference Retrieval, Route Planning, and Counterfactual Physical Deduction. Our benchmark comprises 3,052 high-quality video instances with 4,993 question-answer pairs, constructed via a scalable, visually-grounded pipeline combining advanced model generation with rigorous human verification. Through a comprehensive evaluation of 20 state-of-the-art MLLMs, we uncover significant limitations, revealing that while models demonstrate proficiency in surface-level perception, they exhibit distinct performance drops in MSR tasks, frequently suffering from spatial disorientation and hallucination during multi-step deductions. To mitigate these shortcomings and empower models with stronger MSR capabilities, we further curate MSR-9K, a specialized instruction-tuning dataset, and fine-tune Qwen-VL, achieving a +7.82% absolute improvement on Video-MSR. Our results underscore the efficacy of multi-hop spatial instruction data and establish Video-MSR as a vital foundation for future research. The code and data will be available at https://github.com/ruiz-nju/Video-MSR.

Wang Chen, Guanqiang Qi, Weikang Li et al.

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating external knowledge, but existing approaches indiscriminately trigger retrieval and rely on single-path evidence construction, often introducing noise and limiting performance gains. In this work, we propose Decide Then Retrieve (DTR), a training-free framework that adaptively determines when retrieval is necessary and how external information should be selected. DTR leverages generation uncertainty to guide retrieval triggering and introduces a dual-path retrieval mechanism with adaptive information selection to better handle sparse and ambiguous queries. Extensive experiments across five open-domain QA benchmarks, multiple model scales, and different retrievers demonstrate that DTR consistently improves EM and F1 over standard RAG and strong retrieval-enhanced baselines, while reducing unnecessary retrievals. The code and data used in this paper are available at https://github.com/ChenWangHKU/DTR.

Zhaoxi Zhang, Yitong Duan, Yanzhi Zhang et al. · baidu, tsinghua

Locating files and functions requiring modification in large software repositories is challenging due to their scale and structural complexity. Existing LLM-based methods typically treat this as a repository-level retrieval task and rely on multiple auxiliary tools, which often overlook code execution logic and complicate model control. We propose RepoNavigator, an LLM agent equipped with a single execution-aware tool: jumping to the definition of an invoked symbol. This unified design reflects the actual flow of code execution while simplifying tool manipulation. RepoNavigator is trained end-to-end via Reinforcement Learning (RL) directly from a base pretrained model, without relying on closed-source distillation. Experiments demonstrate that RL-trained RepoNavigator achieves state-of-the-art performance, with the 7B model outperforming 14B baselines, the 14B model surpassing 32B competitors, and the 32B model exceeding closed-source models such as GPT-5 on most metrics. These results confirm that integrating a single, structurally grounded tool with RL training provides an efficient and scalable solution for repository-level issue localization.

Mingyang Liao, Yichen Wan, shuchen wu et al. · baidu, tsinghua

LLM-based role-playing has rapidly improved in fidelity, yet stronger adherence to persona constraints commonly increases vulnerability to jailbreak attacks, especially for risky or negative personas. Most prior work mitigates this issue with training-time solutions (e.g., data curation or alignment-oriented regularization). However, these approaches are costly to maintain as personas and attack strategies evolve, can degrade in-character behavior, and are typically infeasible for frontier closed-weight LLMs. We propose a training-free Dual-Cycle Adversarial Self-Evolution framework with two coupled cycles. A Persona-Targeted Attacker Cycle synthesizes progressively stronger jailbreak prompts, while a Role-Playing Defender Cycle distills observed failures into a hierarchical knowledge base of (i) global safety rules, (ii) persona-grounded constraints, and (iii) safe in-character exemplars. At inference time, the Defender retrieves and composes structured knowledge from this hierarchy to guide generation, producing responses that remain faithful to the target persona while satisfying safety constraints. Extensive experiments across multiple proprietary LLMs show consistent gains over strong baselines on both role fidelity and jailbreak resistance, and robust generalization to unseen personas and attack prompts.

Xin Shen, Zhishu Jiang, Jiaye Yang et al. · baidu, tsinghua

Immersive conversational systems in production face a persistent trade-off between responsiveness and long-horizon task capability. Real-time interaction is achievable for lightweight turns, but requests involving planning and tool invocation (e.g., search and media generation) produce heavy-tail execution latency that degrades turn-taking, persona consistency, and user trust. To address this challenge, we propose DuCCAE (Conversation while Collaboration with Augmentation and Evolution), a hybrid engine for immersive conversation deployed within Baidu Search, serving millions of users. DuCCAE decouples real-time response generation from asynchronous agentic execution and synchronizes them via a shared state that maintains session context and execution traces, enabling asynchronous results to be integrated back into the ongoing dialogue. The system orchestrates five subsystems-Info, Conversation, Collaboration, Augmentation, and Evolution-to support multi-agent collaboration and continuous improvement. We evaluate DuCCAE through a comprehensive framework that combines offline benchmarking on the Du-Interact dataset and large-scale production evaluation within Baidu Search. Experimental results demonstrate that DuCCAE outperforms strong baselines in agentic execution reliability and dialogue quality while reducing latency to fit strict real-time budgets. Crucially, deployment metrics since June 2025 confirm substantial real-world effectiveness, evidenced by a tripling of Day-7 user retention to 34.2% and a surge in the complex task completion rate to 65.2%. Our hybrid architecture successfully preserves conversational continuity while enabling reliable agentic execution, offering practical guidelines for deploying scalable agentic systems in industrial settings.

Zhipeng Ma, Zheyan Tu, Xinhai Chen et al. · baidu, tsinghua

Trajectory representation learning plays a pivotal role in supporting various downstream tasks. Traditional methods in order to filter the noise in GPS trajectories tend to focus on routing-based methods used to simplify the trajectories. However, this approach ignores the motion details contained in the GPS data, limiting the representation capability of trajectory representation learning. To fill this gap, we propose a novel representation learning framework that Joint GPS and Route Modelling based on self-supervised technology, namely JGRM. We consider GPS trajectory and route as the two modes of a single movement observation and fuse information through inter-modal information interaction. Specifically, we develop two encoders, each tailored to capture representations of route and GPS trajectories respectively. The representations from the two modalities are fed into a shared transformer for inter-modal information interaction. Eventually, we design three self-supervised tasks to train the model. We validate the effectiveness of the proposed method on two real datasets based on extensive experiments. The experimental results demonstrate that JGRM outperforms existing methods in both road segment representation and trajectory representation tasks. Our source code is available at Anonymous Github.

Jiulong Wu, Yucheng Shen, Lingyong Yan et al.

Facial Emotion Analysis (FEA) extends traditional facial emotion recognition by incorporating explainable, fine-grained reasoning. The task integrates three subtasks: emotion recognition, facial Action Unit (AU) recognition, and AU-based emotion reasoning to model affective states jointly. While recent approaches leverage Vision-Language Models (VLMs) and achieve promising results, they face two critical limitations: (1) hallucinated reasoning, where VLMs generate plausible but inaccurate explanations due to insufficient emotion-specific knowledge; and (2) misalignment between emotion reasoning and recognition, caused by fragmented connections between observed facial features and final labels. We propose Facial-R1, a three-stage alignment framework that effectively addresses both challenges with minimal supervision. First, we employ instruction fine-tuning to establish basic emotional reasoning capability. Second, we introduce reinforcement training guided by emotion and AU labels as reward signals, which explicitly aligns the generated reasoning process with the predicted emotion. Third, we design a data synthesis pipeline that iteratively leverages the prior stages to expand the training dataset, enabling scalable self-improvement of the model. Built upon this framework, we introduce FEA-20K, a benchmark dataset comprising 17,737 training and 1,688 test samples with fine-grained emotion analysis annotations. Extensive experiments across eight standard benchmarks demonstrate that Facial-R1 achieves state-of-the-art performance in FEA, with strong generalization and robust interpretability.

Dayu Wang, Jiaye Yang, Weikang Li et al.

Large Language Models (LLMs) often suffer from ''Reasoning Collapse'' on challenging mathematical reasoning tasks, where stochastic sampling produces lexical variations of the same erroneous logic rather than genuine semantic exploration. We observe that failed reasoning traces are often associated with a low-rank bias manifold in the model's hidden-state geometry, which reduces exploration toward corrective solution directions. To address this, we propose Spectral Orthogonal Exploration (SOE), a geometric inference framework under a ''Student Guides Teacher'' paradigm. Instead of using a weak auxiliary agent for imitation, SOE uses it as an orthogonal probe to introduce semantically heterogeneous reasoning signals into the teacher's orthogonal complement of its dominant subspace. This intervention steers the teacher toward more diverse reasoning trajectories and improves exploration beyond standard sampling. Experiments on mathematical benchmarks show that SOE improves average accuracy by 62.4\% and average sampling efficiency by 113.7\% over baseline methods, suggesting that geometric interventions can be effective for mitigating reasoning collapse in mathematical reasoning. We further provide preliminary evidence that SOE is also effective on logic and code generation benchmarks.

Lingyong Yan, Jiulong Wu, Dong Xie et al.

Although recent end-to-end video generation models demonstrate impressive performance in visually oriented content creation, they remain limited in scenarios that require strict logical rigor and precise knowledge representation, such as instructional and educational media. To address this problem, we propose LAVES, a hierarchical LLM-based multi-agent system for generating high-quality instructional videos from educational problems. The LAVES formulates educational video generation as a multi-objective task that simultaneously demands correct step-by-step reasoning, pedagogically coherent narration, semantically faithful visual demonstrations, and precise audio--visual alignment. To address the limitations of prior approaches--including low procedural fidelity, high production cost, and limited controllability--LAVES decomposes the generation workflow into specialized agents coordinated by a central Orchestrating Agent with explicit quality gates and iterative critique mechanisms. Specifically, the Orchestrating Agent supervises a Solution Agent for rigorous problem solving, an Illustration Agent that produces executable visualization codes, and a Narration Agent for learner-oriented instructional scripts. In addition, all outputs from the working agents are subject to semantic critique, rule-based constraints, and tool-based compilation checks. Rather than directly synthesizing pixels, the system constructs a structured executable video script that is deterministically compiled into synchronized visuals and narration using template-driven assembly rules, enabling fully automated end-to-end production without manual editing. In large-scale deployments, LAVES achieves a throughput exceeding one million videos per day, delivering over a 95% reduction in cost compared to current industry-standard approaches while maintaining a high acceptance rate.

Deguo Xia, Weiming Zhang, Xiyan Liu et al. · baidu, tsinghua

An up-to-date city-scale lane-level map is an indispensable infrastructure and a key enabling technology for ensuring the safety and user experience of autonomous driving systems. In industrial scenarios, reliance on manual annotation for map updates creates a critical bottleneck. Lane-level updates require precise change information and must ensure consistency with adjacent data while adhering to strict standards. Traditional methods utilize a three-stage approach-construction, change detection, and updating-which often necessitates manual verification due to accuracy limitations. This results in labor-intensive processes and hampers timely updates. To address these challenges, we propose LDMapNet-U, which implements a new end-to-end paradigm for city-scale lane-level map updating. By reconceptualizing the update task as an end-to-end map generation process grounded in historical map data, we introduce a paradigm shift in map updating that simultaneously generates vectorized maps and change information. To achieve this, a Prior-Map Encoding (PME) module is introduced to effectively encode historical maps, serving as a critical reference for detecting changes. Additionally, we incorporate a novel Instance Change Prediction (ICP) module that learns to predict associations with historical maps. Consequently, LDMapNet-U simultaneously achieves vectorized map element generation and change detection. To demonstrate the superiority and effectiveness of LDMapNet-U, extensive experiments are conducted using large-scale real-world datasets. In addition, LDMapNet-U has been successfully deployed in production at Baidu Maps since April 2024, supporting map updating for over 360 cities and significantly shortening the update cycle from quarterly to weekly. The updated maps serve hundreds of millions of users and are integrated into the autonomous driving systems of several leading vehicle companies.

Feifan Xia, Mingyang Liao, Yuyang Fang et al. · baidu, tsinghua

Traditional parameter-efficient fine-tuning (PEFT) methods such as LoRA are tightly coupled with the base model architecture, which constrains their applicability across heterogeneous pretrained large language models (LLMs). To address this limitation, we introduce Cross-LoRA, a data-free framework for transferring LoRA modules between diverse base models without requiring additional training data. Cross-LoRA consists of two key components: (a) LoRA-Align, which performs subspace alignment between source and target base models through rank-truncated singular value decomposition (SVD) and Frobenius-optimal linear transformation, ensuring compatibility under dimension mismatch; and (b) LoRA-Shift, which applies the aligned subspaces to project source LoRA weight updates into the target model parameter space. Both components are data-free, training-free, and enable lightweight adaptation on a commodity GPU in 20 minutes. Experiments on ARCs, OBOA and HellaSwag show that Cross-LoRA achieves relative gains of up to 5.26% over base models. Across other commonsense reasoning benchmarks, Cross-LoRA maintains performance comparable to that of directly trained LoRA adapters.

Wang Chen, Guanqiang Qi, Weikang Li et al. · baidu, tsinghua

Retrieval-Augmented Generation (RAG) has become a cornerstone technique for enhancing large language models (LLMs) with external knowledge. However, current RAG systems face two critical limitations: (1) they inefficiently retrieve information for every query, including simple questions that could be resolved using the LLM's parametric knowledge alone, and (2) they risk retrieving irrelevant documents when queries contain sparse information signals. To address these gaps, we introduce Parametric-verified Adaptive Information Retrieval and Selection (PAIRS), a training-free framework that integrates parametric and retrieved knowledge to adaptively determine whether to retrieve and how to select external information. Specifically, PAIRS employs a dual-path generation mechanism: First, the LLM produces both a direct answer and a context-augmented answer using self-generated pseudo-context. When these outputs converge, PAIRS bypasses external retrieval entirely, dramatically improving the RAG system's efficiency. For divergent cases, PAIRS activates a dual-path retrieval (DPR) process guided by both the original query and self-generated contextual signals, followed by an Adaptive Information Selection (AIS) module that filters documents through weighted similarity to both sources. This simple yet effective approach can not only enhance efficiency by eliminating unnecessary retrievals but also improve accuracy through contextually guided retrieval and adaptive information selection. Experimental results on six question-answering (QA) benchmarks show that PAIRS reduces retrieval costs by around 25% (triggering for only 75% of queries) while still improving accuracy-achieving +1.1% EM and +1.0% F1 over prior baselines on average.

Wang Chen, Wenhan Yu, Guanqiang Qi et al. · baidu, tsinghua

Retrieval-Augmented Generation (RAG) has become a core paradigm in document question answering tasks. However, existing methods have limitations when dealing with multimodal documents: one category of methods relies on layout analysis and text extraction, which can only utilize explicit text information and struggle to capture images or unstructured content; the other category treats document segmentation as visual input and directly passes it to visual language models (VLMs) for processing, yet it ignores the semantic advantages of text, leading to suboptimal retrieval and generation results. To address these research gaps, we propose the Co-Modality-based RAG (CMRAG) framework, which can simultaneously leverage texts and images for more accurate retrieval and generation. Our framework includes two key components: (1) a Unified Encoding Model (UEM) that projects queries, parsed text, and images into a shared embedding space via triplet-based training, and (2) a Unified Co-Modality-informed Retrieval (UCMR) method that statistically normalizes similarity scores to effectively fuse cross-modal signals. To support research in this direction, we further construct and release a large-scale triplet dataset of (query, text, image) examples. Experiments demonstrate that our proposed framework consistently outperforms single-modality--based RAG in multiple visual document question-answering (VDQA) benchmarks. The findings of this paper show that integrating co-modality information into the RAG framework in a unified manner is an effective approach to improving the performance of complex VDQA systems.

Wenhan Yu, Wang Chen, Guanqiang Qi et al.

Document Visual Question Answering (DocVQA) is a fundamental task for multimodal document understanding and a key testbed for vision language reasoning. However, most existing DocVQA datasets are limited to the page level and lack fine grained spatial grounding, constraining the interpretability and reasoning capability of Vision Language Models (VLMs). To address this gap, we introduce BBox DocVQA a large scale, bounding box grounded dataset designed to enhance spatial reasoning and evidence localization in visual documents. We further present an automated construction pipeline, Segment Judge and Generate, which integrates a segment model for region segmentation, a VLM for semantic judgment, and another advanced VLM for question answer generation, followed by human verification for quality assurance. The resulting dataset contains 3.6 K diverse documents and 32 K QA pairs, encompassing single and multi region as well as single and multi page scenarios. Each QA instance is grounded on explicit bounding boxes, enabling fine grained evaluation of spatial semantic alignment. Benchmarking multiple state of the art VLMs (e.g., GPT 5, Qwen2.5 VL, and InternVL) on BBox DocVQA reveals persistent challenges in spatial grounding and reasoning accuracy. Furthermore, fine tuning on BBox DocVQA substantially improves both bounding box localization and answer generation, validating its effectiveness for enhancing the reasoning ability of VLMs. Our dataset and code will be publicly released to advance research on interpretable and spatially grounded vision language reasoning.

Xin Shen, Rui Zhu, Lei Shen et al.

Understanding fine-grained human hand motion is fundamental to visual perception, embodied intelligence, and multimodal communication. In this work, we propose Fine-grained Finger-level Hand Motion Captioning (FingerCap), which aims to generate textual descriptions that capture detailed finger-level semantics of hand actions. To support this task, we curate FingerCap-40K, a large-scale corpus of 40K paired hand-motion videos and captions spanning two complementary sources: concise instruction-style finger motions and diverse, naturalistic hand-object interactions. To enable effective evaluation, we employ HandJudge, a LLM-based rubric that measures finger-level correctness and motion completeness. Temporal sparsity remains a fundamental bottleneck for current Video-MLLMs, since sparse RGB sampling is insufficient to capture the subtle, high-frequency dynamics underlying fine finger motions. As a simple and compute-friendly remedy, we introduce FiGOP (Finger Group-of-Pictures), which pairs each RGB keyframe with subsequent hand keypoints until the next keyframe. A lightweight temporal encoder converts the keypoints into motion embeddings and integrates them with RGB features. FiGOP adapts the classic GOP concept to finger motion, recovering fine temporal cues without increasing RGB density. Experiments on FingerCap-40K show that strong open- and closed-source Video-MLLMs still struggle with finger-level reasoning, while our FiGOP-augmented model yield consistent gains under HandJudge and human studies.

Feifan Xia, Yuyang Fang, Defang Li et al. · baidu, tsinghua

We present a probabilistic intent modeling framework for large language model (LLM) agents in multi-turn social dialogue. The framework maintains a belief distribution over a partner's latent intentions, initialized from contextual priors and dynamically updated through likelihood estimation after each utterance. The evolving distribution provides additional contextual grounding for the policy, enabling adaptive dialogue strategies under uncertainty. Preliminary experiments in the SOTOPIA environment show consistent improvements: the proposed framework increases the Overall score by 9.0% on SOTOPIA-All and 4.1% on SOTOPIA-Hard compared with the Qwen2.5-7B baseline, and slightly surpasses an oracle agent that directly observes partner intentions. These early results suggest that probabilistic intent modeling can contribute to the development of socially intelligent LLM agents.

Deguo Xia, Weiming Zhang, Xiyan Liu et al.

Generating city-scale lane-level maps faces significant challenges due to the intricate urban environments, such as blurred or absent lane markings. Additionally, a standard lane-level map requires a comprehensive organization of lane groupings, encompassing lane direction, style, boundary, and topology, yet has not been thoroughly examined in prior research. These obstacles result in labor-intensive human annotation and high maintenance costs. This paper overcomes these limitations and presents an industrial-grade solution named DuMapNet that outputs standardized, vectorized map elements and their topology in an end-to-end paradigm. To this end, we propose a group-wise lane prediction (GLP) system that outputs vectorized results of lane groups by meticulously tailoring a transformer-based network. Meanwhile, to enhance generalization in challenging scenarios, such as road wear and occlusions, as well as to improve global consistency, a contextual prompts encoder (CPE) module is proposed, which leverages the predicted results of spatial neighborhoods as contextual information. Extensive experiments conducted on large-scale real-world datasets demonstrate the superiority and effectiveness of DuMapNet. Additionally, DuMap-Net has already been deployed in production at Baidu Maps since June 2023, supporting lane-level map generation tasks for over 360 cities while bringing a 95% reduction in costs. This demonstrates that DuMapNet serves as a practical and cost-effective industrial solution for city-scale lane-level map generation.