论文检索

Ming Li, Nan Zhang, Chenrui Fan, Hong Jiao, Yanbin Fu, Sydney Peters, Qingshu Xu, Robert Lissitz, Tianyi Zhou

While Large Reasoning Models (LRMs) generate extensive chain-of-thought reasoning, we lack a principled framework for understanding how these thoughts are structured. In this paper, we introduce a novel approach by applying Schoenfeld’s Episode Theory, a classic cognitive framework for human mathematical problem-solving, to analyze the reasoning traces of LRMs. We annotated thousands of sentences and paragraphs from model-generated solutions to math problems using seven cognitive labels (e.g., Plan, Implement, Verify). The result is the first publicly available benchmark for the fine-grained analysis of machine reasoning, including a large annotated corpus and detailed annotation guidebooks. Our preliminary analysis reveals distinct patterns in LRM reasoning, such as the transition dynamics between cognitive states. This framework provides a theoretically grounded methodology for interpreting LRM cognition and enables future work on more controllable and transparent reasoning systems.

Xiwen Liang, Min Lin, Weiqi Ruan, Rongtao Xu, Yuecheng Liu, Jiaqi Chen, Bingqian Lin, Yuzheng Zhuang, Xiaodan Liang

Existing vision-language planning methods perform well on short-horizon tasks but struggle with long-horizon reasoning in dynamic environments due to the difficulty of training models to generate high-quality reasoning processes. To address this, we propose Structured Preference Optimization (SPO), a framework that enhances reasoning and action selection for long-horizon task planning through structured evaluation and optimized training. SPO introduces: 1) Structured Preference Evaluation and Optimization, which evaluates reasoning chains across task relevance, historical consistency (as part of textual coherence), and image awareness (alignment with visual observations) to construct high-quality preference pairs; and 2) Curriculum-Guided Progressive Learning, enabling the model to adapt from simple to complex tasks, thereby improving generalization and robustness. To advance research in vision-language long-horizon task planning, we introduce ExtendaBench, a comprehensive benchmark covering 1,509 tasks across VirtualHome and Habitat 2.0, categorized into ultra-short, short, medium, and long tasks. Experimental results demonstrate that SPO significantly improves reasoning quality and final decision accuracy, outperforming prior methods on long-horizon tasks and underscoring the effectiveness of preference-driven optimization in vision-language task planning. Specifically, SPO achieves a +5.98% GCR and +4.68% SR improvement in VirtualHome and a +3.30% GCR and +2.11% SR improvement in Habitat over the best-performing baselines.

Yubin Ge, Salvatore Romeo, Jason Cai, Monica Sunkara, Yi Zhang

Despite the rapid advancements in LLM agents, they still face the challenge of generating meaningful reflections due to inadequate error analysis and a reliance on rare successful trajectories, especially in complex tasks. In this work, we propose SAMULE, a new framework for self-learning agents powered by a retrospective language model that is trained based on Multi-Level Reflection Synthesis. It first synthesizes high-quality reflections across three complementary levels: Single-Trajectory Learning (micro-level) for detailed error correction; Intra-Task Learning (meso-level) to build error taxonomies across multiple trials of the same task, and Inter-Task Learning (macro-level) to extract transferable insights based on same typed errors from diverse task failures. Then we fine-tune a language model serving as the retrospective model to generate reflections during inference. We further extend our framework to interactive settings through a foresight-based reflection mechanism, enabling agents to proactively reflect and adapt during user interactions by comparing predicted and actual responses. Extensive experiments on three challenging benchmarks—TravelPlanner, NATURAL PLAN, and Tau-bench—demonstrate that our approach significantly outperforms reflection-based baselines. Our results highlight the critical role of well-designed reflection synthesis and failure-centric learning in building self-improving LLM agents.

Seungwon Lim, Sungwoong Kim, Jihwan Yu, Sungjae Lee, Jiwan Chung, Youngjae Yu

Escape rooms present a unique cognitive challenge that demands exploration-driven planning: with the sole instruction to escape the room, players must actively search their environment, collecting information, and finding solutions through repeated trial and error. Motivated by this, we introduce VisEscape, a benchmark of 20 virtual escape rooms specifically designed to evaluate AI models under these challenging conditions, where success depends not only on solving isolated puzzles but also on iteratively constructing and refining spatial-temporal knowledge of a dynamically changing environment. On VisEscape, we observe that even state-of-the-art multi-modal models generally fail to escape the rooms, showing considerable variation in their progress and problem-solving approaches. We find that integrating memory management and reasoning contributes to efficient exploration and enables successive hypothesis formulation and testing, thereby leading to significant improvements in dynamic and exploration-driven environments.

Lanxiao Huang, Daksh Dave, Tyler Cody, Peter A. Beling, Ming Jin

Large Language Models (LLMs) have been explored for automating or enhancing penetration testing tasks, but their effectiveness and reliability across diverse attack phases remain open questions. This study presents a comprehensive evaluation of multiple LLM-based agents, ranging from singular to modular designs, across realistic penetration testing scenarios, analyzing their empirical performance and recurring failure patterns. We further investigate the impact of core functional capabilities on agent success, operationalized through five targeted augmentations: Global Context Memory (GCM), Inter-Agent Messaging (IAM), Context-Conditioned Invocation (CCI), Adaptive Planning (AP), and Real-Time Monitoring (RTM). These interventions respectively support the capabilities of Context Coherence & Retention, Inter-Component Coordination & State Management, Tool Usage Accuracy & Selective Execution, Multi-Step Strategic Planning & Error Detection & Recovery, and Real-Time Dynamic Responsiveness. Our findings reveal that while some architectures natively exhibit select properties, targeted augmentations significantly enhance modular agent performance—particularly in complex, multi-step, and real-time penetration testing scenarios.

Bin Deng, Yizhe Feng, Zeming Liu, Qing Wei, Xiangrong Zhu, Shuai Chen, Yuanfang Guo, Yunhong Wang

Although large language models have enhanced automated travel planning abilities, current systems remain misaligned with real-world scenarios. First, they assume users provide explicit queries, while in reality requirements are often implicit. Second, existing solutions ignore diverse environmental factors and user preferences, limiting the feasibility of plans. Third, systems can only generate plans with basic POI arrangements, failing to provide all-in-one plans with rich details. To mitigate these challenges, we construct a novel dataset RETAIL, which supports decision-making for implicit queries while covering explicit queries, both with and without revision needs. It also enables environmental awareness to ensure plan feasibility under real-world scenarios, while incorporating detailed POI information for all-in-one travel plans. Furthermore, we propose a topic-guided multi-agent framework, termed TGMA. Our experiments reveal that even the strongest existing model achieves merely a 1.0% pass rate, indicating real-world travel planning remains extremely challenging. In contrast, TGMA demonstrates substantially improved performance 2.72%, offering promising directions for real-world travel planning.

Yiheng Hu, Xiaoyang Wang, Qing Liu, Xiwei Xu, Qian Fu, Wenjie Zhang, Liming Zhu

Multimodal Multi-hop question answering requires integrating information from diverse sources, such as images and texts, to derive answers. Existing methods typically rely on sequential retrieval and reasoning, where each step builds on the previous output. However, this single-path paradigm makes them vulnerable to errors due to misleading intermediate steps. Moreover, developing multimodal models can be computationally expensive, often requiring extensive training. To address these limitations, we propose a training-free framework guided by an Adaptive Planning Graph, which consists of planning, retrieval and reasoning modules. The planning module analyzes the current state of the Adaptive Planning Graph, determines the next action and where to expand the graph, which enables dynamic and flexible exploration of reasoning paths. To handle retrieval of text to unspecified target modalities, we devise modality-specific strategies that dynamically adapt to distinct data types. Our approach preserves the characteristics of multimodal information without costly task-specific training, enabling seamless integration with up-to-date models. Finally, the experiments on MultimodalQA and WebQA show that our approach matches or outperforms existing models that rely on training.

Wei Yang, Jinwei Xiao, Hongming Zhang, Qingyang Zhang, Yanna Wang, Bo Xu

Recent advancements in Large Language Models (LLMs) have driven growing interest in LLM-based agents for complex planning tasks. To avoid costly agent training, many studies adopted memory mechanism that enhances LLM with offline experiences or online trajectory analysis. However, existing works focus on single-granularity memory derived from dynamic environmental interactions, which are inherently constrained by the quality of the collected experiences. This limitation, in turn, constrain the diversity of knowledge and the flexibility of planning. We propose Coarse-to-Fine Grounded Memory (CFGM), a novel framework that grounds coarse-to-fine memories with LLM, thereby fully leverage them for flexible adaptation to diverse scenarios. CFGM grounds environmental information into coarse-grained focus points to guide experience collection in training tasks, followed by grounding of actionable hybrid-grained tips from each experience. At inference, CFGM retrieves task-relevant experiences and tips to support planning. When facing environmental anomalies, the LLM grounds the current situation into fine-grained key information, enabling flexible self-QA reflection and plan correction. Extensive experiments on AlfWorld, Webshop and ScienceWorld demonstrate that CFGM significantly outperforms competitive baselines and comprehensively optimizes memory-enhanced LLM Agent system.

Wenwen Li, Kangwei Shi, Yidong Chai

With the rapid expansion of large language model (LLM) applications, there is an emerging shift in the role of LLM-based AI chatbots from serving merely as general inquiry tools to acting as professional service agents. However, current studies often overlook a critical aspect of professional service agents: the act of communicating in a manner consistent with their professional identities. This is of particular importance in the healthcare sector, where effective communication with patients is essential for achieving professional goals, such as promoting patient well-being by encouraging healthy behaviors. To bridge this gap, we propose LAPI (LLM-based Agent with a Professional Identity), a novel framework for designing professional service agent tailored for medical question-and-answer (Q&A) services, ensuring alignment with a specific professional identity. Our method includes a theory-guided task planning process that decomposes complex professional tasks into manageable subtasks aligned with professional objectives and a pragmatic entropy method designed to generate professional and ethical responses with low uncertainty. Experiments on various LLMs show that the proposed approach outperforms baseline methods, including few-shot prompting, chain-of-thought prompting, across key metrics such as fluency, naturalness, empathy, patient-centricity, and ROUGE-L scores. Additionally, the ablation study underscores the contribution of each component to the overall effectiveness of the approach.

Dong Wang, Xinghang Li, Zhengshen Zhang, Jirong Liu, Xiao Ma, Hanbo Zhang, Tao Kong, Huaping Liu

We introduce ProcWorld, a large-scale benchmark for partially observable embodied spatial reasoning and long-term planning with large language models (LLM) and vision language models (VLM). ProcWorld features a wide range of challenging embodied navigation and object manipulation tasks, covering 16 task types, 5,000 rooms, and over 10 million evaluation trajectories with diverse data distribution. ProcWorld supports configurable observation modes, ranging from text-only descriptions to vision-only observations. It enables text-based actions to control the agent following language instructions. ProcWorld has presented significant challenges for LLMs and VLMs: (1) active information gathering given partial observations for disambiguation; (2) simultaneous localization and decision-making by tracking the spatio-temporal state-action distribution; (3) constrained reasoning with dynamic states subject to physical reachability. Our extensive evaluation of 15 foundation models and 5 reasoning algorithms (with over 1 million rollouts) indicates larger models perform better. However, ProcWorld remains highly challenging for existing state-of-the-art models and in-context learning methods due to constrained reachability and the need for combinatorial spatial reasoning.

Yong Zhao, Kai Xu, Zhengqiu Zhu, Yue Hu, Zhiheng Zheng, Yingfeng Chen, Yatai Ji, Chen Gao, Yong Li, Jincai Huang

Embodied Question Answering (EQA) has primarily focused on indoor environments, leaving the complexities of urban settings—spanning environment, action, and perception—largely unexplored. To bridge this gap, we introduce CityEQA, a new task where an embodied agent answers open-vocabulary questions through active exploration in dynamic city spaces. To support this task, we present CityEQA-EC, the first benchmark dataset featuring 1,412 human-annotated tasks across six categories, grounded in a realistic 3D urban simulator. Moreover, we propose -Manager-Actor (PMA), a novel agent tailored for CityEQA. PMA enables long-horizon planning and hierarchical task execution: the Planner breaks down the question answering into sub-tasks, the Manager maintains an object-centric cognitive map for spatial reasoning during the process control, and the specialized Actors handle navigation, exploration, and collection sub-tasks. Experiments demonstrate that PMA achieves 60.7% of human-level answering accuracy, significantly outperforming frontier-based baselines. While promising, the performance gap compared to humans highlights the need for enhanced visual reasoning in CityEQA. This work paves the way for future advancements in urban spatial intelligence. Dataset and code are available at https://github.com/tsinghua-fib-lab/CityEQA.git.

Hang Ni, Fan Liu, Xinyu Ma, Lixin Su, Shuaiqiang Wang, Dawei Yin, Hui Xiong, Hao Liu

Large language models (LLMs) have shown promise in automating travel planning, yet they often fall short in addressing nuanced spatiotemporal rationality. While existing benchmarks focus on basic plan validity, they neglect critical aspects such as route efficiency, POI appeal, and real-time adaptability. This paper introduces **TP-RAG**, the first benchmark tailored for retrieval-augmented, spatiotemporal-aware travel planning. Our dataset includes 2,348 real-world travel queries, 85,575 fine-grain annotated POIs, and 18,784 high-quality travel trajectory references sourced from online tourist documents, enabling dynamic and context-aware planning. Through extensive experiments, we reveal that integrating reference trajectories significantly improves spatial efficiency and POI rationality of the travel plan, while challenges persist in universality and robustness due to conflicting references and noisy data. To address these issues, we propose *EvoRAG*, an evolutionary framework that potently synergizes diverse retrieved trajectories with LLMs’ intrinsic reasoning. *EvoRAG* achieves state-of-the-art performance, improving spatiotemporal compliance and reducing commonsense violation compared to ground-up and retrieval-augmented baselines. Our work underscores the potential of hybridizing Web knowledge with LLM-driven optimization, paving the way for more reliable and adaptive travel planning agents.

Nishant Balepur, Matthew Shu, Yoo Yeon Sung, Seraphina Goldfarb-Tarrant, Shi Feng, Fumeng Yang, Rachel Rudinger, Jordan Boyd-Graber

To assist users in complex tasks, LLMs generate plans: step-by-step instructions towards a goal. While alignment methods aim to ensure LLM plans are helpful, they train (RLHF) or evaluate (ChatbotArena) on what users prefer, assuming this reflects what helps them. We test this with Planorama: an interface where 126 users answer 300 multi-step questions with LLM plans. We get 4388 plan executions and 5584 comparisons to measure plan helpfulness (QA success) and user preferences on plans, and recreate the setup in agents and reward models to see if they simulate or prefer what helps users. We expose: 1) user/model preferences and agent success do not accurately predict which plans help users, so common alignment feedback can misalign with helpfulness; 2) this gap is not due to user-specific preferences, as users are similarly successful when using plans they prefer/disprefer; 3) surface-level cues like brevity and question similarity strongly link to preferences, but such biases fail to predict helpfulness. In all, we argue aligning helpful LLMs needs feedback from real user interactions—not just preferences of what looks helpful—so we discuss the plan NLP researchers can execute to solve this problem.

Lars Benedikt Kaesberg, Jan Philip Wahle, Terry Ruas, Bela Gipp

Existing reasoning datasets saturate and fail to test abstract, multi-step problems, especially pathfinding and complex rule constraint satisfaction. We introduce SPaRC (Spatial Pathfinding Reasoning Challenge), a dataset of 1,000 2D grid pathfinding puzzles to evaluate spatial and rule-based reasoning, requiring step-by-step planning with arithmetic and geometric rules. Humans achieve near-perfect accuracy (98.0%; 94.5% on hard puzzles), while the best reasoning models, such as o4-mini, struggle (15.8%; 1.1% on hard puzzles). Models often generate invalid paths (>50% of puzzles for o4-mini), and reasoning tokens reveal they make errors in navigation and spatial logic. Unlike humans, who take longer on hard puzzles, models fail to scale test-time compute with difficulty. Allowing models to make multiple solution attempts improves accuracy, suggesting potential for better spatial reasoning with improved training and efficient test-time scaling methods. SPaRC can be used as a window into models’ spatial reasoning limitations and drive research toward new methods that excel in abstract, multi-step problem-solving.

Xueguan Zhao, Wenpeng Lu, Chaoqun Zheng, Weiyu Zhang, Jiasheng Si, Deyu Zhou

Argumentative essay generation (AEG) is a complex task that requires advanced semantic understanding, logical reasoning, and organized integration of perspectives. Despite showing a promising performance, current efforts often overlook the dynamical and hierarchical nature of structural argumentative planning, and struggle with flexible rhetorical expression, leading to limited argument divergence and rhetorical optimization. Inspired by human debate behavior and Bitzer’s rhetorical situation theory, we propose a debate-driven rhetorical framework for argumentative writing. The uniqueness lies in three aspects: (1) dynamic assesses the divergence of viewpoints and progressively reveals the hierarchical outline of arguments based on a depth-then-breadth paradigm, improving the perspective divergence within argumentation; (2) simulates human debate through iterative defender-attacker interactions, improving the logical coherence of arguments; (3) incorporates Bitzer’s rhetorical situation theory to flexibly select appropriate rhetorical techniques, enabling the rhetorical expression. Experiments on four benchmarks validate that our approach significantly improves logical depth, argumentative diversity, and rhetorical persuasiveness over existing state-of-the-art models.

Pengyu Zeng, Jun Yin, Miao Zhang, Yuqin Dai, Jizhizi Li, ZhanXiang Jin, Shuai Lu

In recent years, architectural design automation has made significant progress, but the complexity of open-world environments continues to make residential design a challenging task, often requiring experienced architects to perform multiple iterations and human-computer interactions. Therefore, assisting ordinary users in navigating these complex environments to generate and edit residential design is crucial. In this paper, we present the CARD framework, which leverages a system of specialized cross-modal agents to adapt to complex open-world environments. The framework includes a point-based cross-modal information representation (CMI-P) that encodes the geometry and spatial relationships of residential rooms, a cross-modal residential generation model, supported by our customized Text2FloorEdit model, that acts as the lead designer to create standardized floor plans, and an embedded expert knowledge base for evaluating whether the designs meet user requirements and residential codes, providing feedback accordingly. Finally, a 3D rendering module assists users in visualizing and understanding the layout. CARD enables cross-modal residential generation from free-text input, empowering users to adapt to complex environments without requiring specialized expertise.

Shuodi Liu, Yingzhuo Liu, Zi Wang, Yusheng Wang, Huijia Wu, Liuyu Xiang, Zhaofeng He

Large language models (LLMs) have demonstrated remarkable reasoning and planning capabilities, driving extensive research into task decomposition. Existing task decomposition methods focus primarily on memory, tool usage, and feedback mechanisms, achieving notable success in specific domains, but they often overlook the trade-off between performance and cost. In this study, we first conduct a comprehensive investigation on task decomposition, identifying six categorization schemes. Then, we perform an empirical analysis of three factors that influence the performance and cost of task decomposition: categories of approaches, characteristics of tasks, and configuration of decomposition and execution models, uncovering three critical insights and summarizing a set of practical principles. Building on this analysis, we propose the Select-Then-Decompose strategy, which establishes a closed-loop problem-solving process composed of three stages: selection, execution, and verification. This strategy dynamically selects the most suitable decomposition approach based on task characteristics and enhances the reliability of the results through a verification module. Comprehensive evaluations across multiple benchmarks show that the Select-Then-Decompose consistently lies on the Pareto frontier, demonstrating an optimal balance between performance and cost. Our code is publicly available at https://github.com/summervvind/Select-Then-Decompose.

Hao Sun, Zile Qiao, Bo Wang, Guoxin Chen, Yingyan Hou, Yong Jiang, Pengjun Xie, Fei Huang, Yan Zhang

Retrieval-Augmented Generation (RAG) systems have emerged as a pivotal methodology for enhancing Large Language Models (LLMs) through the dynamic integration of external knowledge. To further improve RAG’s flexibility, Agentic RAG introduces autonomous agents into the workflow. However, Agentic RAG faces several challenges:(1) the success of each step depends on both high-quality planning and accurate search,(2) the lack of supervision for intermediate reasoning steps, and(3) the exponentially large candidate space for planning and searching.To address these challenges, we propose DecoupleSearch, a novel framework that decouples planning and search processes using dual value models, enabling independent optimization of plan reasoning and search grounding. Our approach constructs a reasoning tree, where each node represents planning and search steps. We leverage Monte Carlo Tree Search to assess the quality of each step. During inference, Hierarchical Beam Search iteratively refines planning and search candidates with dual value models. Extensive experiments across policy models of varying parameter sizes, demonstrate the effectiveness of our method.

Lingyao Li, Dawei Li, Zhenhui Ou, Xiaoran Xu, Jingxiao Liu, Zihui Ma, Runlong Yu, Min Deng

Efficient simulation is essential for enhancing proactive preparedness for sudden-onset disasters such as earthquakes. Recent advancements in large language models (LLMs) as world models show promise in simulating complex scenarios. This study examines multiple LLMs to proactively estimate perceived earthquake impacts. Leveraging multimodal datasets including geospatial, socioeconomic, building, and street-level imagery data, our framework generates Modified Mercalli Intensity (MMI) predictions at zip code and county scales. Evaluations on the 2014 Napa and 2019 Ridgecrest earthquakes using USGS “Did You Feel It? (DYFI)” reports demonstrate significant alignment, as evidenced by high correlation of 0.88 and low RMSE of 0.77 as compared to real reports at the zip code level. Techniques such as RAG and ICL can improve simulation performance, while visual inputs notably enhance accuracy compared to structured numerical data alone. These findings show the promise of LLMs in simulating disaster impacts that can help strengthen pre-event planning.

Dayu Yang, Tianyang Liu, Daoan Zhang, Antoine Simoulin, Xiaoyi Liu, Yuwei Cao, Zhaopu Teng, Xin Qian, Grey Yang, Jiebo Luo 等

Code and reasoning recently exhibit a mutually reinforcing relationship in large language models (LLMs): Code is abstract, modular, highly structured and has strong logic, guiding reasoning in training and inference. While reasoning translates high-level goals into small executable steps, enable more sophisticated code intellignece, solving real-world challenging software development problems. In this study, we examine how code serves as a structured medium for enhancing reasoning - providing verifiable execution paths, enforcing logical decomposition, and enabling runtime validation, and how advances in reasoning have transformed code intelligence from basic completion to sophisticated agent - enabling models to tackle complex software engineering tasks through deliberate planning and systematic debugging. Finally, we identify key challenges and propose future research directions may deepen the synergy, ultimately advancing LLM performance in both complex reasoning and code intelligence.