Large language models (LLMs) have demonstrated strong reasoning, planning, and communication abilities, enabling them to operate as autonomous agents in open environments. While single-agent systems remain limited in adaptability and coordination, recent progress has shifted attention toward multi-agent systems (MAS) composed of interacting LLMs that pursue cooperative, competitive, or mixed objectives. This emerging paradigm provides a powerful testbed for studying social dynamics and strategic behaviors among intelligent agents. However, current research remains fragmented and lacks a unifying theoretical foundation. To address this gap, we present a comprehensive survey of LLM-based multi-agent systems through a game-theoretic lens. By organizing existing studies around the four key elements of game theory: players, strategies, payoffs, and information, we establish a systematic framework for understanding, comparing, and guiding future research on the design and analysis of LLM-based MAS.
Recent advances in large language models (LLMs) allow agents to represent actions as executable code, offering greater expressivity than traditional tool-calling. However, real-world tasks often demand both strategic planning and detailed implementation. Using a single agent for both leads to context pollution from debugging traces and intermediate failures, impairing long-horizon performance. We propose CodeDelegator, a multi-agent framework that separates planning from implementation via role specialization. A persistent Delegator maintains strategic oversight by decomposing tasks, writing specifications, and monitoring progress without executing code. For each sub-task, a new Coder agent is instantiated with a clean context containing only its specification, shielding it from prior failures. To coordinate between agents, we introduce Ephemeral-Persistent State Separation (EPSS), which isolates each Coder's execution state while preserving global coherence, preventing debugging traces from polluting the Delegator's context. Experiments on various benchmarks demonstrate the effectiveness of CodeDelegator across diverse scenarios.
We introduce Strategic Doctrine Language Models (sdLM), a learning-system framework for multi-document strategic reasoning with doctrinal consistency constraints and calibrated uncertainty. The approach combines multi-document attention, temporal encoding, and a doctrine-consistency layer to improve long-horizon forecasting and plan plausibility while reducing severe doctrinal violations. We evaluate sdLM using (i) expert-panel scoring of strategic scenarios (N=47), (ii) doctrine consistency on 336 doctrine publications (12,847 statements), and (iii) geopolitical forecasting on 127 historical counterfactuals (1945-2020) across 12-60 month horizons. Across these benchmarks, sdLM achieves higher strategic quality and better calibration than strong general-purpose LLM baselines, and remains competitive with human experts on long-horizon judgments. We further report ablations, scaling trends, and deployment-oriented performance/latency characteristics to clarify which components drive improvements and how they translate to operational settings.
Optimizing the advertiser's cumulative value of winning impressions under budget constraints poses a complex challenge in online advertising, under the paradigm of AI-Generated Bidding (AIGB). Advertisers often have personalized objectives but limited historical interaction data, resulting in few-shot scenarios where traditional reinforcement learning (RL) methods struggle to perform effectively. Large Language Models (LLMs) offer a promising alternative for AIGB by leveraging their in-context learning capabilities to generalize from limited data. However, they lack the numerical precision required for fine-grained optimization. To address this limitation, we introduce GRPO-Adaptive, an efficient LLM post-training strategy that enhances both reasoning and numerical precision by dynamically updating the reference policy during training. Built upon this foundation, we further propose DARA, a novel dual-phase framework that decomposes the decision-making process into two stages: a few-shot reasoner that generates initial plans via in-context prompting, and a fine-grained optimizer that refines these plans using feedback-driven reasoning. This separation allows DARA to combine LLMs' in-context learning strengths with precise adaptability required by AIGB tasks. Extensive experiments on both real-world and synthetic data environments demonstrate that our approach consistently outperforms existing baselines in terms of cumulative advertiser value under budget constraints.
Recent progress in large language models (LLMs) has shown that reasoning improves when intermediate thoughts are externalized into explicit workspaces, such as chain-of-thought traces or tool-augmented reasoning. Yet, visual language models (VLMs) lack an analogous mechanism for spatial reasoning, limiting their ability to generate images that accurately reflect geometric relations, object identities, and compositional intent. We introduce the concept of a spatial scratchpad -- a 3D reasoning substrate that bridges linguistic intent and image synthesis. Given a text prompt, our framework parses subjects and background elements, instantiates them as editable 3D meshes, and employs agentic scene planning for placement, orientation, and viewpoint selection. The resulting 3D arrangement is rendered back into the image domain with identity-preserving cues, enabling the VLM to generate spatially consistent and visually coherent outputs. Unlike prior 2D layout-based methods, our approach supports intuitive 3D edits that propagate reliably into final images. Empirically, it achieves a 32% improvement in text alignment on GenAI-Bench, demonstrating the benefit of explicit 3D reasoning for precise, controllable image generation. Our results highlight a new paradigm for vision-language models that deliberate not only in language, but also in space. Code and visualizations at https://oindrilasaha.github.io/3DScratchpad/
Recent work shows that fine-tuned Large Language Models (LLMs) can achieve high valid plan rates on PDDL planning tasks. However, it remains unclear whether this reflects transferable planning competence or domain-specific memorization. In this work, we fine-tune a 1.7B-parameter LLM on 40,000 domain-problem-plan tuples from 10 IPC 2023 domains, and evaluate both in-domain and cross-domain generalization. While the model reaches 82.9% valid plan rate in in-domain conditions, it achieves 0% on two unseen domains. To analyze this failure, we introduce three diagnostic interventions, namely (i) instance-wise symbol anonymization, (ii) compact plan serialization, and (iii) verifier-reward fine-tuning using the VAL validator as a success-focused reinforcement signal. Symbol anonymization and compact serialization cause significant performance drops despite preserving plan semantics, thus revealing strong sensitivity to surface representations. Verifier-reward fine-tuning reaches performance saturation in half the supervised training epochs, but does not improve cross-domain generalization. For the explored configurations, in-domain performance plateaus around 80%, while cross-domain performance collapses, suggesting that our fine-tuned model relies heavily on domain-specific patterns rather than transferable planning competence in this setting. Our results highlight a persistent generalization gap in LLM-based planning and provide diagnostic tools for studying its causes.
The integration of agentic AI, powered by large language models (LLMs) with autonomous reasoning, planning, and execution, into unmanned aerial vehicle (UAV) swarms opens new operational possibilities and brings the vision of the Internet of Drones closer to reality. However, infrastructure constraints, dynamic environments, and the computational demands of multi-agent coordination limit real-world deployment in high-risk scenarios such as wildfires and disaster response. This paper investigates the integration of LLM-based agentic AI and edge computing to realize scalable and resilient autonomy in UAV swarms. We first discuss three architectures for supporting UAV swarms - standalone, edge-enabled, and edge-cloud hybrid deployment - each optimized for varying autonomy and connectivity levels. Then, a use case for wildfire search and rescue (SAR) is designed to demonstrate the efficiency of the edge-enabled architecture, enabling high SAR coverage, reduced mission completion times, and a higher level of autonomy compared to traditional approaches. Finally, we highlight open challenges in integrating LLMs and edge computing for mission-critical UAV-swarm applications.
Embodied Artificial Intelligence (AI) has recently attracted significant attention as it bridges AI with the physical world. Modern embodied AI systems often combine a Large Language Model (LLM)-based planner for high-level task planning and a reinforcement learning (RL)-based controller for low-level action generation, enabling embodied agents to tackle complex tasks in real-world environments. However, deploying embodied agents remains challenging due to their high computation requirements, especially for battery-powered local devices. Although techniques like lowering operating voltage can improve energy efficiency, they can introduce bit errors and result in task failures. In this work, we propose CREATE, a general design principle that leverages heterogeneous resilience at different layers for synergistic energy-reliability co-optimization. For the first time, we conduct a comprehensive error injection study on modern embodied AI systems and observe an inherent but heterogeneous fault tolerance. Building upon these insights, we develop an anomaly detection and clearance mechanism at the circuit level to eliminate outlier errors. At the model level, we propose a weight-rotation-enhanced planning algorithm to improve the fault tolerance of the LLM-based planner. Furthermore, we introduce an application-level technique, autonomy-adaptive voltage scaling, to dynamically adjust the operating voltage of the controllers. The voltage scaling circuit is co-designed to enable online voltage adjustment. Extensive experiments demonstrate that without compromising task quality, CREATE achieves 40.6% computational energy savings on average over nominal-voltage baselines and 35.0% over prior-art techniques. This further leads to 29.5% to 37.3% chip-level energy savings and approximately a 15% to 30% improvement in battery life.
Robots are expected to play a major role in the future construction industry but face challenges due to high costs and difficulty adapting to dynamic tasks. This study explores the potential of foundation models to enhance the adaptability and generalizability of task planning in construction robots. Four models are proposed and implemented using lightweight, open-source large language models (LLMs) and vision language models (VLMs). These models include one single agent and three multi-agent teams that collaborate to create robot action plans. The models are evaluated across three construction roles: Painter, Safety Inspector, and Floor Tiling. Results show that the four-agent team outperforms the state-of-the-art GPT-4o in most metrics while being ten times more cost-effective. Additionally, teams with three and four agents demonstrate the improved generalizability. By discussing how agent behaviors influence outputs, this study enhances the understanding of AI teams and supports future research in diverse unstructured environments beyond construction.
Large language models (LLMs) have shown strong capabilities in multi-step decision-making, planning and actions, and are increasingly integrated into various real-world applications. It is concerning whether their strong problem-solving abilities may be misused for crimes. To address this gap, we propose VirtualCrime, a sandbox simulation framework based on a three-agent system to evaluate the criminal capabilities of models. Specifically, this framework consists of an attacker agent acting as the leader of a criminal team, a judge agent determining the outcome of each action, and a world manager agent updating the environment state and entities. Furthermore, we design 40 diverse crime tasks within this framework, covering 11 maps and 13 crime objectives such as theft, robbery, kidnapping, and riot. We also introduce a human player baseline for reference to better interpret the performance of LLM agents. We evaluate 8 strong LLMs and find (1) All agents in the simulation environment compliantly generate detailed plans and execute intelligent crime processes, with some achieving relatively high success rates; (2) In some cases, agents take severe action that inflicts harm to NPCs to achieve their goals. Our work highlights the need for safety alignment when deploying agentic AI in real-world settings.
Large Language Models (LLMs) such as ChatGPT promise revolutionary capabilities for Sixth-Generation (6G) wireless networks but their massive computational requirements and tendency to generate technically incorrect information create deployment barriers. In this work, we introduce MAINTAINED: autonomous artificial intelligence agent for wireless network deployment. Instead of encoding domain knowledge within model parameters, our approach orchestrates specialized computational tools for geographic analysis, signal propagation modeling, and network optimization. In a real-world case study, MAINTAINED outperforms state-of-the-art LLMs including ChatGPT-4o, Claude Sonnet 4, and DeepSeek-R1 by up to 100-fold in verified performance metrics while requiring less computational resources. This paradigm shift, moving from relying on parametric knowledge towards externalizing domain knowledge into verifiable computational tools, eliminates hallucination in technical specifications and enables edge-deployable Artificial Intelligence (AI) for wireless communications.
Understanding traveler behavior and accurately predicting travel mode choice are at the heart of transportation planning and policy-making. This study proposes TransMode-LLM, an innovative framework that integrates statistical methods with LLM-based techniques to predict travel modes from travel survey data. The framework operates through three phases: (1) statistical analysis identifies key behavioral features, (2) natural language encoding transforms structured data into contextual descriptions, and (3) LLM adaptation predicts travel mode through multiple learning paradigms including zero-shot and one/few-shot learning and domain-enhanced prompting. We evaluate TransMode-LLM using both general-purpose models (GPT-4o, GPT-4o-mini) and reasoning-focused models (o3-mini, o4-mini) with varying sample sizes on real-world travel survey data. Extensive experiment results demonstrate that the LLM-based approach achieves competitive accuracy compared to state-of-the-art baseline classifiers models. Moreover, few-shot learning significantly improves prediction accuracy, with models like o3-mini showing consistent improvements of up to 42.9\% with 5 provided examples. However, domain-enhanced prompting shows divergent effects across LLM architectures. In detail, it is helpful to improve performance for general-purpose models with GPT-4o achieving improvements of 2.27% to 12.50%. However, for reasoning-oriented models (o3-mini, o4-mini), domain knowledge enhancement does not universally improve performance. This study advances the application of LLMs in travel behavior modeling, providing promising and valuable insights for both academic research and transportation policy-making in the future.
Existing dynamic Theory of Mind (ToM) benchmarks mostly place language models in a passive role: the model reads a sequence of connected scenarios and reports what people believe, feel, intend, and do as these states change. In real social interaction, ToM is also used for action: a speaker plans what to say in order to shift another person's mental-state trajectory toward a goal. We introduce SocialMindChange, a benchmark that moves from tracking minds to changing minds in social interaction. Each instance defines a social context with 4 characters and five connected scenes. The model plays one character and generates dialogue across the five scenes to reach the target while remaining consistent with the evolving states of all participants. SocialMindChange also includes selected higher-order states. Using a structured four-step framework, we construct 1,200 social contexts, covering 6000 scenarios and over 90,000 questions, each validated for realism and quality. Evaluations on ten state-of-the-art LLMs show that their average performance is 54.2% below human performance. This gap suggests that current LLMs still struggle to maintain and change mental-state representations across long, linked interactions.
Navigation agents powered by large language models (LLMs) convert natural language instructions into executable plans and actions. Compared to text-based applications, their security is far more critical: a successful prompt injection attack does not just alter outputs but can directly misguide physical navigation, leading to unsafe routes, mission failure, or real-world harm. Despite this high-stakes setting, the vulnerability of navigation agents to prompt injection remains largely unexplored. In this paper, we propose PINA, an adaptive prompt optimization framework tailored to navigation agents under black-box, long-context, and action-executable constraints. Experiments on indoor and outdoor navigation agents show that PINA achieves high attack success rates with an average ASR of 87.5%, surpasses all baselines, and remains robust under ablation and adaptive-attack conditions. This work provides the first systematic investigation of prompt injection attacks in navigation and highlights their urgent security implications for embodied LLM agents.
Simulated environments play an essential role in embodied AI, functionally analogous to test cases in software engineering. However, existing environment generation methods often emphasize visual realism (e.g., object diversity and layout coherence), overlooking a crucial aspect: logical diversity from the testing perspective. This limits the comprehensive evaluation of agent adaptability and planning robustness in distinct simulated environments. To bridge this gap, we propose LogicEnvGen, a novel method driven by Large Language Models (LLMs) that adopts a top-down paradigm to generate logically diverse simulated environments as test cases for agents. Given an agent task, LogicEnvGen first analyzes its execution logic to construct decision-tree-structured behavior plans and then synthesizes a set of logical trajectories. Subsequently, it adopts a heuristic algorithm to refine the trajectory set, reducing redundant simulation. For each logical trajectory, which represents a potential task situation, LogicEnvGen correspondingly instantiates a concrete environment. Notably, it employs constraint solving for physical plausibility. Furthermore, we introduce LogicEnvEval, a novel benchmark comprising four quantitative metrics for environment evaluation. Experimental results verify the lack of logical diversity in baselines and demonstrate that LogicEnvGen achieves 1.04-2.61x greater diversity, significantly improving the performance in revealing agent faults by 4.00%-68.00%.
The emergence of LLMs has catalyzed a paradigm shift in autonomous agent development, enabling systems capable of reasoning, planning, and executing complex multi-step tasks. However, existing agent frameworks often suffer from architectural rigidity, vendor lock-in, and prohibitive complexity that impedes rapid prototyping and deployment. This paper presents AgentForge, a lightweight, open-source Python framework designed to democratize the construction of LLM-driven autonomous agents through a principled modular architecture. AgentForge introduces three key innovations: (1) a composable skill abstraction that enables fine-grained task decomposition with formally defined input-output contracts, (2) a unified LLM backend interface supporting seamless switching between cloud-based APIs and local inference engines, and (3) a declarative YAML-based configuration system that separates agent logic from implementation details. We formalize the skill composition mechanism as a directed acyclic graph (DAG) and prove its expressiveness for representing arbitrary sequential and parallel task workflows. Comprehensive experimental evaluation across four benchmark scenarios demonstrates that AgentForge achieves competitive task completion rates while reducing development time by 62% compared to LangChain and 78% compared to direct API integration. Latency measurements confirm sub-100ms orchestration overhead, rendering the framework suitable for real-time applications. The modular design facilitates extension: we demonstrate the integration of six built-in skills and provide comprehensive documentation for custom skill development. AgentForge addresses a critical gap in the LLM agent ecosystem by providing researchers and practitioners with a production-ready foundation for constructing, evaluating, and deploying autonomous agents without sacrificing flexibility or performance.
Across the social and medical sciences, researchers recognize that specifying planned research activities (i.e., 'registration') prior to the commencement of research has benefits for both the transparency and rigour of science. Despite this, evidence suggests that study registrations frequently go unexamined, minimizing their effectiveness. In a way this is no surprise: manually checking registrations against papers is labour- and time-intensive, requiring careful reading across formats and expertise across domains. The advent of AI unlocks new possibilities in facilitating this activity. We present RegCheck, a modular LLM-assisted tool designed to help researchers, reviewers, and editors from across scientific disciplines compare study registrations with their corresponding papers. Importantly, RegCheck keeps human expertise and judgement in the loop by (i) ensuring that users are the ones who determine which features should be compared, and (ii) presenting the most relevant text associated with each feature to the user, facilitating (rather than replacing) human discrepancy judgements. RegCheck also generates shareable reports with unique RegCheck IDs, enabling them to be easily shared and verified by other users. RegCheck is designed to be adaptable across scientific domains, as well as registration and publication formats. In this paper we provide an overview of the motivation, workflow, and design principles of RegCheck, and we discuss its potential as an extensible infrastructure for reproducible science with an example use case.
Current Large Language Models (LLMs) exhibit a critical modal disconnect: they possess vast semantic knowledge but lack the procedural grounding to respect the immutable laws of the physical world. Consequently, while these agents implicitly function as world models, their simulations often suffer from physical hallucinations-generating plans that are logically sound but physically unexecutable. Existing alignment strategies predominantly rely on resource-intensive training or fine-tuning, which attempt to compress dynamic environmental rules into static model parameters. However, such parametric encapsulation is inherently rigid, struggling to adapt to the open-ended variability of physical dynamics without continuous, costly retraining. To bridge this gap, we introduce WorldMind, a framework that autonomously constructs a symbolic World Knowledge Repository by synthesizing environmental feedback. Specifically, it unifies Process Experience to enforce physical feasibility via prediction errors and Goal Experience to guide task optimality through successful trajectories. Experiments on EB-ALFRED and EB-Habitat demonstrate that WorldMind achieves superior performance compared to baselines with remarkable cross-model and cross-environment transferability.
Maritime port inspection plays a critical role in ensuring safety, regulatory compliance, and operational efficiency in complex maritime environments. However, existing inspection methods often rely on manual operations and conventional computer vision techniques that lack scalability and contextual understanding. This study introduces a novel integrated engineering framework that utilizes the synergy between Large Language Models (LLMs) and Vision Language Models (VLMs) to enable autonomous maritime port inspection using cooperative aerial and surface robotic platforms. The proposed framework replaces traditional state-machine mission planners with LLM-driven symbolic planning and improved perception pipelines through VLM-based semantic inspection, enabling context-aware and adaptive monitoring. The LLM module translates natural language mission instructions into executable symbolic plans with dependency graphs that encode operational constraints and ensure safe UAV-USV coordination. Meanwhile, the VLM module performs real-time semantic inspection and compliance assessment, generating structured reports with contextual reasoning. The framework was validated using the extended MBZIRC Maritime Simulator with realistic port infrastructure and further assessed through real-world robotic inspection trials. The lightweight on-board design ensures suitability for resource-constrained maritime platforms, advancing the development of intelligent, autonomous inspection systems. Project resources (code and videos) can be found here: https://github.com/Muhayyuddin/llm-vlm-fusion-port-inspection
The accelerating growth of the scientific literature makes it increasingly difficult for researchers to track new advances through manual reading alone. Recent progress in large language models (LLMs) has therefore spurred interest in autonomous agents that can read scientific papers and extract task-relevant information. However, most existing approaches rely either on heavily engineered prompting or on a conventional SFT-RL training pipeline, both of which often lead to excessive and low-yield exploration. Drawing inspiration from cognitive science, we propose PaperCompass, a framework that mitigates these issues by separating high-level planning from fine-grained execution. PaperCompass first drafts an explicit plan that outlines the intended sequence of actions, and then performs detailed reasoning to instantiate each step by selecting the parameters for the corresponding function calls. To train such behavior, we introduce Draft-and-Follow Policy Optimization (DFPO), a tailored RL method that jointly optimizes both the draft plan and the final solution. DFPO can be viewed as a lightweight form of hierarchical reinforcement learning, aimed at narrowing the `knowing-doing' gap in LLMs. We provide a theoretical analysis that establishes DFPO's favorable optimization properties, supporting a stable and reliable training process. Experiments on paper-based question answering (Paper-QA) benchmarks show that PaperCompass improves efficiency over strong baselines without sacrificing performance, achieving results comparable to much larger models.
Understanding and classifying query intents can improve retrieval effectiveness by helping align search results with the motivations behind user queries. However, existing intent taxonomies are typically derived from system log data and capture mostly isolated information needs, while the broader task context often remains unaddressed. This limitation becomes increasingly relevant as interactions with Large Language Models (LLMs) expand user expectations from simple query answering toward comprehensive task support, for example, with purchasing decisions or in travel planning. At the same time, current LLMs still struggle to fully interpret complex and multifaceted tasks. To address this gap, we argue for a stronger task-based perspective on query intent. Drawing on a grounded-theory-based interview study with airport information clerks, we present a taxonomy of task-based information request intents that bridges the gap between traditional query-focused approaches and the emerging demands of AI-driven task-oriented search.
Equipping Large Language Models (LLMs) with external tools enables them to solve complex real-world problems. However, the robustness of existing methods remains a critical challenge when confronting novel or evolving tools. Existing trajectory-centric paradigms primarily rely on memorizing static solution paths during training, which limits the ability of LLMs to generalize tool usage to newly introduced or previously unseen tools. In this paper, we propose ToolMaster, a framework that shifts tool use from imitating golden tool-calling trajectories to actively learning tool usage through interaction with the environment. To optimize LLMs for tool planning and invocation, ToolMaster adopts a trial-and-execution paradigm, which trains LLMs to first imitate teacher-generated trajectories containing explicit tool trials and self-correction, followed by reinforcement learning to coordinate the trial and execution phases jointly. This process enables agents to autonomously explore correct tool usage by actively interacting with environments and forming experiential knowledge that benefits tool execution. Experimental results demonstrate that ToolMaster significantly outperforms existing baselines in terms of generalization and robustness across unseen or unfamiliar tools. All code and data are available at https://github.com/NEUIR/ToolMaster.
GPU code optimization is a key performance bottleneck for HPC workloads as well as large-model training and inference. Although compiler optimizations and hand-written kernels can partially alleviate this issue, achieving near-hardware-limit performance still relies heavily on manual code refactoring and parameter tuning. Recent progress in LLM-agent-based kernel generation and optimization has been reported, yet many approaches primarily focus on direct code rewriting, where parameter choices are often implicit and hard to control, or require human intervention, leading to unstable performance gains. This paper introduces a template-based rewriting layer on top of an agent-driven iterative loop: kernels are semantically refactored into explicitly parameterizable templates, and template parameters are then optimized via search-based autotuning, yielding more stable and higher-quality speedups. Experiments on a set of real-world kernels demonstrate speedups exceeding 3x in the best case. We extract representative CUDA kernels from SGLang as evaluation targets; the proposed agentic tuner iteratively performs templating, testing, analysis, and planning, and leverages profiling feedback to execute constrained parameter search under hardware resource limits. Compared to agent-only direct rewriting, the template-plus-search design significantly reduces the randomness of iterative optimization, making the process more interpretable and enabling a more systematic approach toward high-performance configurations. The proposed method can be further extended to OpenCL, HIP, and other backends to deliver automated performance optimization for real production workloads.
Artificial Intelligence is moving from models that only generate text to Agentic AI, where systems behave as autonomous entities that can perceive, reason, plan, and act. Large Language Models (LLMs) are no longer used only as passive knowledge engines but as cognitive controllers that combine memory, tool use, and feedback from their environment to pursue extended goals. This shift already supports the automation of complex workflows in software engineering, scientific discovery, and web navigation, yet the variety of emerging designs, from simple single loop agents to hierarchical multi agent systems, makes the landscape hard to navigate. In this paper, we investigate architectures and propose a unified taxonomy that breaks agents into Perception, Brain, Planning, Action, Tool Use, and Collaboration. We use this lens to describe the move from linear reasoning procedures to native inference time reasoning models, and the transition from fixed API calls to open standards like the Model Context Protocol (MCP) and Native Computer Use. We also group the environments in which these agents operate, including digital operating systems, embodied robotics, and other specialized domains, and we review current evaluation practices. Finally, we highlight open challenges, such as hallucination in action, infinite loops, and prompt injection, and outline future research directions toward more robust and reliable autonomous systems.
Reasoning is a fundamental cognitive process underlying inference, problem-solving, and decision-making. While large language models (LLMs) demonstrate strong reasoning capabilities in closed-world settings, they struggle in open-ended and dynamic environments. Agentic reasoning marks a paradigm shift by reframing LLMs as autonomous agents that plan, act, and learn through continual interaction. In this survey, we organize agentic reasoning along three complementary dimensions. First, we characterize environmental dynamics through three layers: foundational agentic reasoning, which establishes core single-agent capabilities including planning, tool use, and search in stable environments; self-evolving agentic reasoning, which studies how agents refine these capabilities through feedback, memory, and adaptation; and collective multi-agent reasoning, which extends intelligence to collaborative settings involving coordination, knowledge sharing, and shared goals. Across these layers, we distinguish in-context reasoning, which scales test-time interaction through structured orchestration, from post-training reasoning, which optimizes behaviors via reinforcement learning and supervised fine-tuning. We further review representative agentic reasoning frameworks across real-world applications and benchmarks, including science, robotics, healthcare, autonomous research, and mathematics. This survey synthesizes agentic reasoning methods into a unified roadmap bridging thought and action, and outlines open challenges and future directions, including personalization, long-horizon interaction, world modeling, scalable multi-agent training, and governance for real-world deployment.
Recent advances in LLM-based multi-agent systems (MAS) show that workflows composed of multiple LLM agents with distinct roles, tools, and communication patterns can outperform single-LLM baselines on complex tasks. However, most frameworks are homogeneous, where all agents share the same base LLM and differ only in prompts, tools, and positions in the workflow. This raises the question of whether such workflows can be simulated by a single agent through multi-turn conversations. We investigate this across seven benchmarks spanning coding, mathematics, general question answering, domain-specific reasoning, and real-world planning and tool use. Our results show that a single agent can reach the performance of homogeneous workflows with an efficiency advantage from KV cache reuse, and can even match the performance of an automatically optimized heterogeneous workflow. Building on this finding, we propose \textbf{OneFlow}, an algorithm that automatically tailors workflows for single-agent execution, reducing inference costs compared to existing automatic multi-agent design frameworks without trading off accuracy. These results position the single-LLM implementation of multi-agent workflows as a strong baseline for MAS research. We also note that single-LLM methods cannot capture heterogeneous workflows due to the lack of KV cache sharing across different LLMs, highlighting future opportunities in developing \textit{truly} heterogeneous multi-agent systems.
The use of Large Language Models (LLMs) for reasoning and planning tasks has drawn increasing attention in Artificial Intelligence research. Despite their remarkable progress, these models still exhibit limitations in multi-step inference scenarios, particularly in mathematical and logical reasoning. We introduce PREGU (Partial Reasoning Guided by Uncertainty). PREGU monitors the entropy of the output distribution during autoregressive generation and halts the process whenever entropy exceeds a defined threshold, signaling uncertainty. From that point, a localized search is performed in the latent space to refine the partial reasoning and select the most coherent answer, using the Soft Reasoning method. Experiments conducted with LLaMA-3-8B, Mistral-7B, and Qwen2-7B across four reasoning benchmarks (GSM8K, GSM-Hard, SVAMP, and StrategyQA) showed performance greater than or similar to Soft Reasoning, indicating that entropy can serve as an effective signal to trigger selective refinement during reasoning.
The safety validation of autonomous robotic vehicles hinges on systematically testing their planning and control stacks against rare, safety-critical scenarios. Mining these long-tail events from massive real-world driving logs is therefore a critical step in the robotic development lifecycle. The goal of the Scenario Mining task is to retrieve useful information to enable targeted re-simulation, regression testing, and failure analysis of the robot's decision-making algorithms. RefAV, introduced by the Argoverse team, is an end-to-end framework that uses large language models (LLMs) to spatially and temporally localize scenarios described in natural language. However, this process performs retrieval on trajectory labels, ignoring the direct connection between natural language and raw RGB images, which runs counter to the intuition of video retrieval; it also depends on the quality of upstream 3D object detection and tracking. Further, inaccuracies in trajectory data lead to inaccuracies in downstream spatial and temporal localization. To address these issues, we propose Robust Scenario Mining for Robotic Autonomy from Coarse to Fine (SMc2f), a coarse-to-fine pipeline that employs vision-language models (VLMs) for coarse image-text filtering, builds a database of successful mining cases on top of RefAV and automatically retrieves exemplars to few-shot condition the LLM for more robust retrieval, and introduces text-trajectory contrastive learning to pull matched pairs together and push mismatched pairs apart in a shared embedding space, yielding a fine-grained matcher that refines the LLM's candidate trajectories. Experiments on public datasets demonstrate substantial gains in both retrieval quality and efficiency.
Large language models (LLMs) struggle with reasoning over long contexts where relevant information is sparsely distributed. Although plan-and-execute frameworks mitigate this by decomposing tasks into planning and execution, their effectiveness is often limited by unreliable plan generation due to dependence on surface-level cues. Consequently, plans may be based on incorrect assumptions, and once a plan is formed, identifying what went wrong and revising it reliably becomes difficult, limiting the effectiveness of reactive refinement. To address this limitation, we propose PPA-Plan, a proactive planning strategy for long-context reasoning that focuses on preventing such failures before plan generation. PPA-Plan identifies potential logical pitfalls and false assumptions, formulates them as negative constraints, and conditions plan generation on explicitly avoiding these constraints. Experiments on long-context QA benchmarks show that executing plans generated by PPA-Plan consistently outperforms existing plan-and-execute methods and direct prompting.
We introduce a unified framework that seamlessly integrates algorithmic recourse, contextual bandits, and large language models (LLMs) to support sequential decision-making in high-stakes settings such as personalized medicine. We first introduce the recourse bandit problem, where a decision-maker must select both a treatment action and a feasible, minimal modification to mutable patient features. To address this problem, we develop the Generalized Linear Recourse Bandit (GLRB) algorithm. Building on this foundation, we propose LIBRA, a Language Model-Informed Bandit Recourse Algorithm that strategically combines domain knowledge from LLMs with the statistical rigor of bandit learning. LIBRA offers three key guarantees: (i) a warm-start guarantee, showing that LIBRA significantly reduces initial regret when LLM recommendations are near-optimal; (ii) an LLM-effort guarantee, proving that the algorithm consults the LLM only $O(\log^2 T)$ times, where $T$ is the time horizon, ensuring long-term autonomy; and (iii) a robustness guarantee, showing that LIBRA never performs worse than a pure bandit algorithm even when the LLM is unreliable. We further establish matching lower bounds that characterize the fundamental difficulty of the recourse bandit problem and demonstrate the near-optimality of our algorithms. Experiments on synthetic environments and a real hypertension-management case study confirm that GLRB and LIBRA improve regret, treatment quality, and sample efficiency compared with standard contextual bandits and LLM-only benchmarks. Our results highlight the promise of recourse-aware, LLM-assisted bandit algorithms for trustworthy LLM-bandits collaboration in personalized high-stakes decision-making.