The emergence of LLM-based agents represents a paradigm shift in AI, enabling autonomous systems to plan, reason, use tools, and maintain memory while interacting with dynamic environments. This paper provides the first comprehensive survey of evaluation methodologies for these increasingly capable agents. We systematically analyze evaluation benchmarks and frameworks across four critical dimensions: (1) fundamental agent capabilities, including planning, tool use, self-reflection, and memory; (2) application-specific benchmarks for web, software engineering, scientific, and conversational agents; (3) benchmarks for generalist agents; and (4) frameworks for evaluating agents. Our analysis reveals emerging trends, including a shift toward more realistic, challenging evaluations with continuously updated benchmarks. We also identify critical gaps that future research must address-particularly in assessing cost-efficiency, safety, and robustness, and in developing fine-grained, and scalable evaluation methods. This survey maps the rapidly evolving landscape of agent evaluation, reveals the emerging trends in the field, identifies current limitations, and proposes directions for future research.
Large language models (LLMs) have recently transformed from text-based assistants to autonomous agents capable of planning, reasoning, and iteratively improving their actions. While numerical reward signals and verifiers can effectively rank candidate actions, they often provide limited contextual guidance. In contrast, natural language feedback better aligns with the generative capabilities of LLMs, providing richer and more actionable suggestions. However, parsing and implementing this feedback effectively can be challenging for LLM-based agents. In this work, we introduce Critique-Guided Improvement (CGI), a novel two-player framework, comprising an actor model that explores an environment and a critic model that generates detailed nature language feedback. By training the critic to produce fine-grained assessments and actionable revisions, and the actor to utilize these critiques, our approach promotes more robust exploration of alternative strategies while avoiding local optima. Experiments in three interactive environments show that CGI outperforms existing baselines by a substantial margin. Notably, even a small critic model surpasses GPT-4 in feedback quality. The resulting actor achieves state-of-the-art performance, demonstrating the power of explicit iterative guidance to enhance decision-making in LLM-based agents.
In autonomous driving, open-ended question answering often suffers from unreliable evaluations because freeform responses require either complex metrics or subjective human judgment. To address this challenge, we introduce AutoDrive-QA, an automatic pipeline that converts existing driving QA datasets (including DriveLM, NuScenes-QA, and LingoQA) into a structured multiple-choice question (MCQ) format. This benchmark systematically assesses perception, prediction, and planning tasks, providing a standardized and objective evaluation framework. AutoDrive-QA employs an automated pipeline that leverages large language models (LLMs) to generate high-quality, contextually relevant distractors based on domain-specific error patterns commonly found in autonomous driving scenarios. To evaluate both general capabilities and generalization performance, we test the benchmark on three public datasets and conduct zero-shot experiments on an unseen dataset. The zero-shot evaluations reveal that GPT-4V leads with 69.57% accuracy -- achieving 74.94% in Perception, 65.33% in Prediction, and 68.45% in Planning -- demonstrating that while all models excel in Perception, they struggle in Prediction. Consequently, AutoDrive-QA establishes a rigorous, unbiased standard for integrating and evaluating different vision-language models across various autonomous driving datasets, thereby improving generalization in this field. We release all the codes in the AutoDrive-QA GitHub Repository.
The integration of large language models (LLMs) into robotic task planning has unlocked better reasoning capabilities for complex, long-horizon workflows. However, ensuring safety in LLM-driven plans remains a critical challenge, as these models often prioritize task completion over risk mitigation. This paper introduces SAFER (Safety-Aware Framework for Execution in Robotics), a multi-LLM framework designed to embed safety awareness into robotic task planning. SAFER employs a Safety Agent that operates alongside the primary task planner, providing safety feedback. Additionally, we introduce LLM-as-a-Judge, a novel metric leveraging LLMs as evaluators to quantify safety violations within generated task plans. Our framework integrates safety feedback at multiple stages of execution, enabling real-time risk assessment, proactive error correction, and transparent safety evaluation. We also integrate a control framework using Control Barrier Functions (CBFs) to ensure safety guarantees within SAFER's task planning. We evaluated SAFER against state-of-the-art LLM planners on complex long-horizon tasks involving heterogeneous robotic agents, demonstrating its effectiveness in reducing safety violations while maintaining task efficiency. We also verify the task planner and safety planner through actual hardware experiments involving multiple robots and a human.
Large Language Models (LLMs) have demonstrated remarkable text generation capabilities, and recent advances in training paradigms have led to breakthroughs in their reasoning performance. In this work, we investigate how the reasoning effort of such models scales with problem complexity. We use the infinitely scalable Tents puzzle, which has a known linear-time solution, to analyze this scaling behavior. Our results show that reasoning effort scales with problem size, but only up to a critical problem complexity. Beyond this threshold, the reasoning effort does not continue to increase, and may even decrease. This observation highlights a critical limitation in the logical coherence of current LLMs as problem complexity increases, and underscores the need for strategies to improve reasoning scalability. Furthermore, our results reveal significant performance differences between current state-of-the-art reasoning models when faced with increasingly complex logical puzzles.
While Large Language Models (LLMs) excel at reasoning on text and Vision-Language Models (VLMs) are highly effective for visual perception, applying those models for visual instruction-based planning remains a widely open problem. In this paper, we introduce VIPER, a novel framework for multimodal instruction-based planning that integrates VLM-based perception with LLM-based reasoning. Our approach uses a modular pipeline where a frozen VLM generates textual descriptions of image observations, which are then processed by an LLM policy to predict actions based on the task goal. We fine-tune the reasoning module using behavioral cloning and reinforcement learning, improving our agent's decision-making capabilities. Experiments on the ALFWorld benchmark show that VIPER significantly outperforms state-of-the-art visual instruction-based planners while narrowing the gap with purely text-based oracles. By leveraging text as an intermediate representation, VIPER also enhances explainability, paving the way for a fine-grained analysis of perception and reasoning components.
This paper addresses the high demand in advanced intelligent robot navigation for a more holistic understanding of spatial environments, by introducing a novel system that harnesses the capabilities of Large Language Models (LLMs) to construct hierarchical 3D Scene Graphs (3DSGs) for indoor scenarios. The proposed framework constructs 3DSGs consisting of a fundamental layer with rich metric-semantic information, an object layer featuring precise point-cloud representation of object nodes as well as visual descriptors, and higher layers of room, floor, and building nodes. Thanks to the innovative application of LLMs, not only object nodes but also nodes of higher layers, e.g., room nodes, are annotated in an intelligent and accurate manner. A polling mechanism for room classification using LLMs is proposed to enhance the accuracy and reliability of the room node annotation. Thorough numerical experiments demonstrate the system's ability to integrate semantic descriptions with geometric data, creating an accurate and comprehensive representation of the environment instrumental for context-aware navigation and task planning.
Mixed-precision computing, a widely applied technique in AI, offers a larger trade-off space between accuracy and efficiency. The recent purposed Mixed-Precision Over-the-Air Federated Learning (MP-OTA-FL) enables clients to operate at appropriate precision levels based on their heterogeneous hardware, taking advantages of the larger trade-off space while covering the quantization overheads in the mixed-precision modulation scheme for the OTA aggregation process. A key to further exploring the potential of the MP-OTA-FL framework is the optimization of client precision levels. The choice of precision level hinges on multifaceted factors including hardware capability, potential client contribution, and user satisfaction, among which factors can be difficult to define or quantify. In this paper, we propose a RAG-based User Profiling for precision planning framework that integrates retrieval-augmented LLMs and dynamic client profiling to optimize satisfaction and contributions. This includes a hybrid interface for gathering device/user insights and an RAG database storing historical quantization decisions with feedback. Experiments show that our method boosts satisfaction, energy savings, and global model accuracy in MP-OTA-FL systems.
A Barrier-Free GeoQA Portal: Enhancing Geospatial Data Accessibility with a Multi-Agent LLM Framework Geoportals are vital for accessing and analyzing geospatial data, promoting open spatial data sharing and online geo-information management. Designed with GIS-like interaction and layered visualization, they often challenge non-expert users with complex functionalities and overlapping layers that obscure spatial relationships. We propose a GeoQA Portal using a multi-agent Large Language Model framework for seamless natural language interaction with geospatial data. Complex queries are broken into subtasks handled by specialized agents, retrieving relevant geographic data efficiently. Task plans are shown to users, boosting transparency. The portal supports default and custom data inputs for flexibility. Semantic search via word vector similarity aids data retrieval despite imperfect terms. Case studies, evaluations, and user tests confirm its effectiveness for non-experts, bridging GIS complexity and public access, and offering an intuitive solution for future geoportals.
Planning a trip into a potentially unsafe area is a difficult task. We conducted a formative study on travelers' information needs, finding that most of them turn to search engines for trip planning. Search engines, however, fail to provide easily interpretable results adapted to the context and personal information needs of a traveler. Large language models (LLMs) create new possibilities for providing personalized travel safety advice. To explore this idea, we developed DangerMaps, a mapping system that assists its users in researching the safety of an urban travel destination, whether it is pre-travel or on-location. DangerMaps plots safety ratings onto a map and provides explanations on demand. This late breaking work specifically emphasizes the challenges of designing real-world applications with large language models. We provide a detailed description of our approach to prompt design and highlight future areas of research.
Compression is at the heart of intelligence. A theoretically optimal way to compress any sequence of data is to find the shortest program that outputs that sequence and then halts. However, such 'Kolmogorov compression' is uncomputable, and code generating LLMs struggle to approximate this theoretical ideal, as it requires reasoning, planning and search capabilities beyond those of current models. In this work, we introduce the KoLMogorov-Test (KT), a compression-as-intelligence test for code generating LLMs. In KT a model is presented with a sequence of data at inference time, and asked to generate the shortest program that produces the sequence. We identify several benefits of KT for both evaluation and training: an essentially infinite number of problem instances of varying difficulty is readily available, strong baselines already exist, the evaluation metric (compression) cannot be gamed, and pretraining data contamination is highly unlikely. To evaluate current models, we use audio, text, and DNA data, as well as sequences produced by random synthetic programs. Current flagship models perform poorly - both GPT4-o and Llama-3.1-405B struggle on our natural and synthetic sequences. On our synthetic distribution, we are able to train code generation models with lower compression rates than previous approaches. Moreover, we show that gains on synthetic data generalize poorly to real data, suggesting that new innovations are necessary for additional gains on KT.
The aspiration of the Vision-and-Language Navigation (VLN) task has long been to develop an embodied agent with robust adaptability, capable of seamlessly transferring its navigation capabilities across various tasks. Despite remarkable advancements in recent years, most methods necessitate dataset-specific training, thereby lacking the capability to generalize across diverse datasets encompassing distinct types of instructions. Large language models (LLMs) have demonstrated exceptional reasoning and generalization abilities, exhibiting immense potential in robot action planning. In this paper, we propose FlexVLN, an innovative hierarchical approach to VLN that integrates the fundamental navigation ability of a supervised-learning-based Instruction Follower with the robust generalization ability of the LLM Planner, enabling effective generalization across diverse VLN datasets. Moreover, a verification mechanism and a multi-model integration mechanism are proposed to mitigate potential hallucinations by the LLM Planner and enhance execution accuracy of the Instruction Follower. We take REVERIE, SOON, and CVDN-target as out-of-domain datasets for assessing generalization ability. The generalization performance of FlexVLN surpasses that of all the previous methods to a large extent.
With the increasing prevalence of online learning, adapting education to diverse learner needs remains a persistent challenge. Recent advancements in artificial intelligence (AI), particularly large language models (LLMs), promise powerful tools and capabilities to enhance personalized learning in online educational environments. In this work, we explore how LLMs can improve personalized learning experiences by catering to individual user needs toward enhancing the overall quality of online education. We designed personalization guidelines based on the growing literature on personalized learning to ground LLMs in generating tailored learning plans. To operationalize these guidelines, we implemented LearnMate, an LLM-based system that generates personalized learning plans and provides users with real-time learning support. We discuss the implications and future directions of this work, aiming to move beyond the traditional one-size-fits-all approach by integrating LLM-based personalized support into online learning environments.
As requirements drift with rapid iterations, agile development becomes the dominant paradigm. Goal-driven Requirements Elicitation (RE) is a pivotal yet challenging task in agile project development due to its heavy tangling with adaptive planning and efficient collaboration. Recently, AI agents have shown promising ability in supporting requirements analysis by saving significant time and effort for stakeholders. However, current research mainly focuses on functional RE, and research works have not been reported bridging the long journey from goal to user stories. Moreover, considering the cost of LLM facilities and the need for data and idea protection, privately hosted small-sized LLM should be further utilized in RE. To address these challenges, we propose Goal2Story, a multi-agent fleet that adopts the Impact Mapping (IM) framework while merely using cost-effective sLLMs for goal-driven RE. Moreover, we introduce a StorySeek dataset that contains over 1,000 user stories (USs) with corresponding goals and project context information, as well as the semi-automatic dataset construction method. For evaluation, we proposed two metrics: Factuality Hit Rate (FHR) to measure consistency between the generated USs with the dataset and Quality And Consistency Evaluation (QuACE) to evaluate the quality of the generated USs. Experimental results demonstrate that Goal2Story outperforms the baseline performance of the Super-Agent adopting powerful LLMs, while also showcasing the performance improvements in key metrics brought by CoT and Agent Profile to Goal2Story, as well as its exploration in identifying latent needs.
We introduce a novel large language model (LLM)-driven agent framework, which iteratively refines queries and filters contextual evidence by leveraging dynamically evolving knowledge. A defining feature of the system is its decoupling of external sources from an internal knowledge cache that is progressively updated to guide both query generation and evidence selection. This design mitigates bias-reinforcement loops and enables dynamic, trackable search exploration paths, thereby optimizing the trade-off between exploring diverse information and maintaining accuracy through autonomous agent decision-making. Our approach is evaluated on a broad range of open-domain question answering benchmarks, including multi-step tasks that mirror real-world scenarios where integrating information from multiple sources is critical, especially given the vulnerabilities of LLMs that lack explicit reasoning or planning capabilities. The results show that the proposed system not only outperforms single-step baselines regardless of task difficulty but also, compared to conventional iterative retrieval methods, demonstrates pronounced advantages in complex tasks through precise evidence-based reasoning and enhanced efficiency. The proposed system supports both competitive and collaborative sharing of updated context, enabling multi-agent extension. The benefits of multi-agent configurations become especially prominent as task difficulty increases. The number of convergence steps scales with task difficulty, suggesting cost-effective scalability.
Relational database-driven data analysis (RDB-DA) report generation, which aims to generate data analysis reports after querying relational databases, has been widely applied in fields such as finance and healthcare. Typically, these tasks are manually completed by data scientists, making the process very labor-intensive and showing a clear need for automation. Although existing methods (e.g., Table QA or Text-to-SQL) have been proposed to reduce human dependency, they cannot handle complex analytical tasks that require multi-step reasoning, cross-table associations, and synthesizing insights into reports. Moreover, there is no dataset available for developing automatic RDB-DA report generation. To fill this gap, this paper proposes an LLM agent system for RDB-DA report generation tasks, dubbed DAgent; moreover, we construct a benchmark for automatic data analysis report generation, which includes a new dataset DA-Dataset and evaluation metrics. DAgent integrates planning, tools, and memory modules to decompose natural language questions into logically independent sub-queries, accurately retrieve key information from relational databases, and generate analytical reports that meet the requirements of completeness, correctness, and conciseness through multi-step reasoning and effective data integration. Experimental analysis on the DA-Dataset demonstrates that DAgent's superiority in retrieval performance and analysis report generation quality, showcasing its strong potential for tackling complex database analysis report generation tasks.
Inpatient pathways demand complex clinical decision-making based on comprehensive patient information, posing critical challenges for clinicians. Despite advancements in large language models (LLMs) in medical applications, limited research focused on artificial intelligence (AI) inpatient pathways systems, due to the lack of large-scale inpatient datasets. Moreover, existing medical benchmarks typically concentrated on medical question-answering and examinations, ignoring the multifaceted nature of clinical decision-making in inpatient settings. To address these gaps, we first developed the Inpatient Pathway Decision Support (IPDS) benchmark from the MIMIC-IV database, encompassing 51,274 cases across nine triage departments and 17 major disease categories alongside 16 standardized treatment options. Then, we proposed the Multi-Agent Inpatient Pathways (MAP) framework to accomplish inpatient pathways with three clinical agents, including a triage agent managing the patient admission, a diagnosis agent serving as the primary decision maker at the department, and a treatment agent providing treatment plans. Additionally, our MAP framework includes a chief agent overseeing the inpatient pathways to guide and promote these three clinician agents. Extensive experiments showed our MAP improved the diagnosis accuracy by 25.10% compared to the state-of-the-art LLM HuatuoGPT2-13B. It is worth noting that our MAP demonstrated significant clinical compliance, outperforming three board-certified clinicians by 10%-12%, establishing a foundation for inpatient pathways systems.
With the rapid development of Large Language Models (LLMs), LLM-based agents have been widely adopted in various fields, becoming essential for autonomous decision-making and interactive tasks. However, current work typically relies on prompt design or fine-tuning strategies applied to vanilla LLMs, which often leads to limited effectiveness or suboptimal performance in complex agent-related environments. Although LLM optimization techniques can improve model performance across many general tasks, they lack specialized optimization towards critical agent functionalities such as long-term planning, dynamic environmental interaction, and complex decision-making. Although numerous recent studies have explored various strategies to optimize LLM-based agents for complex agent tasks, a systematic review summarizing and comparing these methods from a holistic perspective is still lacking. In this survey, we provide a comprehensive review of LLM-based agent optimization approaches, categorizing them into parameter-driven and parameter-free methods. We first focus on parameter-driven optimization, covering fine-tuning-based optimization, reinforcement learning-based optimization, and hybrid strategies, analyzing key aspects such as trajectory data construction, fine-tuning techniques, reward function design, and optimization algorithms. Additionally, we briefly discuss parameter-free strategies that optimize agent behavior through prompt engineering and external knowledge retrieval. Finally, we summarize the datasets and benchmarks used for evaluation and tuning, review key applications of LLM-based agents, and discuss major challenges and promising future directions. Our repository for related references is available at https://github.com/YoungDubbyDu/LLM-Agent-Optimization.
Reasoning and strategic behavior in social interactions is a hallmark of intelligence. This form of reasoning is significantly more sophisticated than isolated planning or reasoning tasks in static settings (e.g., math problem solving). In this paper, we present Strategic Planning, Interaction, and Negotiation (SPIN-Bench), a new multi-domain evaluation designed to measure the intelligence of strategic planning and social reasoning. While many existing benchmarks focus on narrow planning or single-agent reasoning, SPIN-Bench combines classical PDDL tasks, competitive board games, cooperative card games, and multi-agent negotiation scenarios in one unified framework. The framework includes both a benchmark as well as an arena to simulate and evaluate the variety of social settings to test reasoning and strategic behavior of AI agents. We formulate the benchmark SPIN-Bench by systematically varying action spaces, state complexity, and the number of interacting agents to simulate a variety of social settings where success depends on not only methodical and step-wise decision making, but also conceptual inference of other (adversarial or cooperative) participants. Our experiments reveal that while contemporary LLMs handle basic fact retrieval and short-range planning reasonably well, they encounter significant performance bottlenecks in tasks requiring deep multi-hop reasoning over large state spaces and socially adept coordination under uncertainty. We envision SPIN-Bench as a catalyst for future research on robust multi-agent planning, social reasoning, and human--AI teaming. Project Website: https://spinbench.github.io/
Safe, agile, and socially compliant multi-robot navigation in cluttered and constrained environments remains a critical challenge. This is especially difficult with self-interested agents in decentralized settings, where there is no central authority to resolve conflicts induced by spatial symmetry. We address this challenge by proposing a novel approach, GameChat, which facilitates safe, agile, and deadlock-free navigation for both cooperative and self-interested agents. Key to our approach is the use of natural language communication to resolve conflicts, enabling agents to prioritize more urgent tasks and break spatial symmetry in a socially optimal manner. Our algorithm ensures subgame perfect equilibrium, preventing agents from deviating from agreed-upon behaviors and supporting cooperation. Furthermore, we guarantee safety through control barrier functions and preserve agility by minimizing disruptions to agents' planned trajectories. We evaluate GameChat in simulated environments with doorways and intersections. The results show that even in the worst case, GameChat reduces the time for all agents to reach their goals by over 35% from a naive baseline and by over 20% from SMG-CBF in the intersection scenario, while doubling the rate of ensuring the agent with a higher priority task reaches the goal first, from 50% (equivalent to random chance) to a 100% perfect performance at maximizing social welfare.
Unmanned Surface Vessels (USVs) are essential for various maritime operations. USV mission planning approach offers autonomous solutions for monitoring, surveillance, and logistics. Existing approaches, which are based on static methods, struggle to adapt to dynamic environments, leading to suboptimal performance, higher costs, and increased risk of failure. This paper introduces a novel mission planning framework that uses Large Language Models (LLMs), such as GPT-4, to address these challenges. LLMs are proficient at understanding natural language commands, executing symbolic reasoning, and flexibly adjusting to changing situations. Our approach integrates LLMs into maritime mission planning to bridge the gap between high-level human instructions and executable plans, allowing real-time adaptation to environmental changes and unforeseen obstacles. In addition, feedback from low-level controllers is utilized to refine symbolic mission plans, ensuring robustness and adaptability. This framework improves the robustness and effectiveness of USV operations by integrating the power of symbolic planning with the reasoning abilities of LLMs. In addition, it simplifies the mission specification, allowing operators to focus on high-level objectives without requiring complex programming. The simulation results validate the proposed approach, demonstrating its ability to optimize mission execution while seamlessly adapting to dynamic maritime conditions.
Large Language Models (LLMs) have advanced autonomous agents' planning and decision-making, yet they struggle with complex tasks requiring diverse expertise and multi-step reasoning. Multi-Agent Debate (MAD) systems, introduced in NLP research, address this gap by enabling structured debates among LLM-based agents to refine solutions iteratively. MAD promotes divergent thinking through role-specific agents, dynamic interactions, and structured decision-making. Recognizing parallels between Software Engineering (SE) and collaborative human problem-solving, this study investigates MAD's effectiveness on two SE tasks. We adapt MAD systems from NLP, analyze agent interactions to assess consensus-building and iterative refinement, and propose two enhancements targeting observed weaknesses. Our findings show that structured debate and collaboration improve problem-solving and yield strong performance in some cases, highlighting MAD's potential for SE automation while identifying areas for exploration.
Recent LLM-based agent frameworks have demonstrated impressive capabilities in task delegation and workflow orchestration, but face significant challenges in maintaining context awareness and ensuring planning consistency. This paper presents SagaLLM, a structured multi-agent framework that addresses four fundamental limitations in current LLM approaches: inadequate self-validation, context narrowing, lacking transaction properties, and insufficient inter-agent coordination. By implementing specialized context management agents and validation protocols, SagaLLM preserves critical constraints and state information throughout complex planning processes, enabling robust and consistent decision-making even during disruptions. We evaluate our approach using selected problems from the REALM benchmark, focusing on sequential and reactive planning scenarios that challenge both context retention and adaptive reasoning. Our experiments with state-of-the-art LLMs, Claude 3.7, DeepSeek R1, GPT-4o, and GPT-o1, demonstrate that while these models exhibit impressive reasoning capabilities, they struggle with maintaining global constraint awareness during complex planning tasks, particularly when adapting to unexpected changes. In contrast, the distributed cognitive architecture of SagaLLM shows significant improvements in planning consistency, constraint enforcement, and adaptation to disruptions in various scenarios.
Human reasoning relies on constructing and manipulating mental models-simplified internal representations of situations that we use to understand and solve problems. Conceptual diagrams (for example, sketches drawn by humans to aid reasoning) externalize these mental models, abstracting irrelevant details to efficiently capture relational and spatial information. In contrast, Large Language Models (LLMs) and Large Multimodal Models (LMMs) predominantly reason through textual representations, limiting their effectiveness in complex multi-step combinatorial and planning tasks. In this paper, we propose a zero-shot fully automatic framework that enables LMMs to reason through multiple chains of self-generated intermediate conceptual diagrams, significantly enhancing their combinatorial planning capabilities. Our approach does not require any human initialization beyond a natural language description of the task. It integrates both textual and diagrammatic reasoning within an optimized graph-of-thought inference framework, enhanced by beam search and depth-wise backtracking. Evaluated on multiple challenging PDDL planning domains, our method substantially improves GPT-4o's performance (for example, from 35.5% to 90.2% in Blocksworld). On more difficult planning domains with solution depths up to 40, our approach outperforms even the o1-preview reasoning model (for example, over 13% improvement in Parking). These results highlight the value of conceptual diagrams as a complementary reasoning medium in LMMs.
Large language models (LLMs) are used in chatbots or AI assistants to hold conversations with a human user. In such applications, the quality (e.g., user engagement, safety) of a conversation is important and can only be exactly known at the end of the conversation. To maximize its expected quality, conversation planning reasons about the stochastic transitions within a conversation to select the optimal LLM response at each turn. Existing simulation-based conversation planning algorithms typically select the optimal response by simulating future conversations with a large number of LLM queries at every turn. However, this process is extremely time-consuming and hence impractical for real-time conversations. This paper presents a novel approach called Semantic space COnversation Planning with improved Efficiency (SCOPE) that exploits the dense semantic representation of conversations to perform conversation planning efficiently. In particular, SCOPE models the stochastic transitions in conversation semantics and their associated rewards to plan entirely within the semantic space. This allows us to select the optimal LLM response at every conversation turn without needing additional LLM queries for simulation. As a result, SCOPE can perform conversation planning 70 times faster than conventional simulation-based planning algorithms when applied to a wide variety of conversation starters and two reward functions seen in the real world, yet achieving a higher reward within a practical planning budget. Our code can be found at: https://github.com/chenzhiliang94/convo-plan-SCOPE.
The rise of Large Reasoning Models (LRMs) signifies a paradigm shift toward advanced computational reasoning. Yet, this progress disrupts traditional agent frameworks, traditionally anchored by execution-oriented Large Language Models (LLMs). To explore this transformation, we propose the LaRMA framework, encompassing nine tasks across Tool Usage, Plan Design, and Problem Solving, assessed with three top LLMs (e.g., Claude3.5-sonnet) and five leading LRMs (e.g., DeepSeek-R1). Our findings address four research questions: LRMs surpass LLMs in reasoning-intensive tasks like Plan Design, leveraging iterative reflection for superior outcomes; LLMs excel in execution-driven tasks such as Tool Usage, prioritizing efficiency; hybrid LLM-LRM configurations, pairing LLMs as actors with LRMs as reflectors, optimize agent performance by blending execution speed with reasoning depth; and LRMs' enhanced reasoning incurs higher computational costs, prolonged processing, and behavioral challenges, including overthinking and fact-ignoring tendencies. This study fosters deeper inquiry into LRMs' balance of deep thinking and overthinking, laying a critical foundation for future agent design advancements.
The adoption of Large Language Models (LLMs) is rapidly expanding across various tasks that involve inherent graphical structures. Graphs are integral to a wide range of applications, including motion planning for autonomous vehicles, social networks, scene understanding, and knowledge graphs. Many problems, even those not initially perceived as graph-based, can be effectively addressed through graph theory. However, when applied to these tasks, LLMs often encounter challenges, such as hallucinations and mathematical inaccuracies. To overcome these limitations, we propose Graph-Grounded LLMs, a system that improves LLM performance on graph-related tasks by integrating a graph library through function calls. By grounding LLMs in this manner, we demonstrate significant reductions in hallucinations and improved mathematical accuracy in solving graph-based problems, as evidenced by the performance on the NLGraph benchmark. Finally, we showcase a disaster rescue application where the Graph-Grounded LLM acts as a decision-support system.
Text-to-image models like stable diffusion and DALLE-3 still struggle with multi-turn image editing. We decompose such a task as an agentic workflow (path) of tool use that addresses a sequence of subtasks by AI tools of varying costs. Conventional search algorithms require expensive exploration to find tool paths. While large language models (LLMs) possess prior knowledge of subtask planning, they may lack accurate estimations of capabilities and costs of tools to determine which to apply in each subtask. Can we combine the strengths of both LLMs and graph search to find cost-efficient tool paths? We propose a three-stage approach "CoSTA*" that leverages LLMs to create a subtask tree, which helps prune a graph of AI tools for the given task, and then conducts A* search on the small subgraph to find a tool path. To better balance the total cost and quality, CoSTA* combines both metrics of each tool on every subtask to guide the A* search. Each subtask's output is then evaluated by a vision-language model (VLM), where a failure will trigger an update of the tool's cost and quality on the subtask. Hence, the A* search can recover from failures quickly to explore other paths. Moreover, CoSTA* can automatically switch between modalities across subtasks for a better cost-quality trade-off. We build a novel benchmark of challenging multi-turn image editing, on which CoSTA* outperforms state-of-the-art image-editing models or agents in terms of both cost and quality, and performs versatile trade-offs upon user preference.
With the rapid advancement of large language models (LLMs) and vision-language models (VLMs), significant progress has been made in developing open-vocabulary robotic manipulation systems. However, many existing approaches overlook the importance of object dynamics, limiting their applicability to more complex, dynamic tasks. In this work, we introduce KUDA, an open-vocabulary manipulation system that integrates dynamics learning and visual prompting through keypoints, leveraging both VLMs and learning-based neural dynamics models. Our key insight is that a keypoint-based target specification is simultaneously interpretable by VLMs and can be efficiently translated into cost functions for model-based planning. Given language instructions and visual observations, KUDA first assigns keypoints to the RGB image and queries the VLM to generate target specifications. These abstract keypoint-based representations are then converted into cost functions, which are optimized using a learned dynamics model to produce robotic trajectories. We evaluate KUDA on a range of manipulation tasks, including free-form language instructions across diverse object categories, multi-object interactions, and deformable or granular objects, demonstrating the effectiveness of our framework. The project page is available at http://kuda-dynamics.github.io.
Operations Research (OR) has been widely applied in various fields such as resource allocation, production planning, and supply chain management. However, addressing real-world OR problems requires OR experts to perform mathematical modeling and programmers to develop solution algorithms. This traditional method, heavily reliant on experts, is costly and has long development cycles, severely limiting the widespread adoption of OR techniques. Few have considered using Artificial Intelligence (AI) to replace professionals to achieve fully automated solutions for OR problems. We propose OR-LLM-Agent, the first AI agent that enables end-to-end automation for solving real-world OR problems. OR-LLM-Agent leverages the Chain-of-Thought (CoT) reasoning capabilities of Large Language Models (LLMs) to translate natural language problem descriptions into formal mathematical models and automatically generate Gurobi solver code. In OR-LLM-Agent, OR-CodeAgent is designed to automate code execution and repair within a sandbox environment, facilitating the derivation of the final solution. Due to the lack of dedicated benchmark datasets for evaluating the automated solving of OR problems, we construct a benchmark dataset comprising 83 real-world OR problems described in natural language. We conduct comparative experiments with state-of-the-art (SOTA) reasoning LLMs, including GPT-o3-mini, DeepSeek-R1, and Gemini 2.0 Flash Thinking. The OR-LLM-Agent achieved the highest pass rate of 100% and the highest solution accuracy of 85%, demonstrating the feasibility of automated OR problem-solving. Data and code have been publicly available at https://github.com/bwz96sco/or_llm_agent.