Failure Analysis (FA) is a highly intricate and knowledge-intensive process. The integration of AI components within the computational infrastructure of FA labs has the potential to automate a variety of tasks, including the detection of non-conformities in images, the retrieval of analogous cases from diverse data sources, and the generation of reports from annotated images. However, as the number of deployed AI models increases, the challenge lies in orchestrating these components into cohesive and efficient workflows that seamlessly integrate with the FA process. This paper investigates the design and implementation of a Large Language Model (LLM)-based Planning Agent (LPA) to assist FA engineers in solving their analysis cases. The LPA integrates LLMs with advanced planning capabilities and external tool utilization, enabling autonomous processing of complex queries, retrieval of relevant data from external systems, and generation of human-readable responses. Evaluation results demonstrate the agent's operational effectiveness and reliability in supporting FA tasks.
Recent advances in Large Reasoning Models (LRMs) trained with Long Chain-of-Thought (Long CoT) reasoning have demonstrated remarkable cross-domain generalization capabilities. However, the underlying mechanisms supporting such transfer remain poorly understood. We hypothesize that cross-domain generalization arises from shared abstract reasoning prototypes -- fundamental reasoning patterns that capture the essence of problems across domains. These prototypes minimize the nuances of the representation, revealing that seemingly diverse tasks are grounded in shared reasoning structures.Based on this hypothesis, we propose ProtoReasoning, a framework that enhances the reasoning ability of LLMs by leveraging scalable and verifiable prototypical representations (Prolog for logical reasoning, PDDL for planning).ProtoReasoning features: (1) an automated prototype construction pipeline that transforms problems into corresponding prototype representations; (2) a comprehensive verification system providing reliable feedback through Prolog/PDDL interpreters; (3) the scalability to synthesize problems arbitrarily within prototype space while ensuring correctness. Extensive experiments show that ProtoReasoning achieves 4.7% improvement over baseline models on logical reasoning (Enigmata-Eval), 6.3% improvement on planning tasks, 4.0% improvement on general reasoning (MMLU) and 1.0% on mathematics (AIME24). Significantly, our ablation studies confirm that learning in prototype space also demonstrates enhanced generalization to structurally similar problems compared to training solely on natural language representations, validating our hypothesis that reasoning prototypes serve as the foundation for generalizable reasoning in large language models.
Large language models (LLMs) are increasingly being adopted as the cognitive core of embodied agents. However, inherited hallucinations, which stem from failures to ground user instructions in the observed physical environment, can lead to navigation errors, such as searching for a refrigerator that does not exist. In this paper, we present the first systematic study of hallucinations in LLM-based embodied agents performing long-horizon tasks under scene-task inconsistencies. Our goal is to understand to what extent hallucinations occur, what types of inconsistencies trigger them, and how current models respond. To achieve these goals, we construct a hallucination probing set by building on an existing benchmark, capable of inducing hallucination rates up to 40x higher than base prompts. Evaluating 12 models across two simulation environments, we find that while models exhibit reasoning, they fail to resolve scene-task inconsistencies-highlighting fundamental limitations in handling infeasible tasks. We also provide actionable insights on ideal model behavior for each scenario, offering guidance for developing more robust and reliable planning strategies.
While knowledge distillation has become a mature field for compressing large language models (LLMs) into smaller ones by aligning their outputs or internal representations, the distillation of LLM-based agents, which involve planning, memory, and tool use, remains relatively underexplored. Existing agent distillation methods typically replay full teacher trajectories or imitate step-by-step teacher tool usage, but they often struggle to train student agents to dynamically plan and act in novel environments. We propose AgentDistill, a novel, training-free agent distillation framework that enables efficient and scalable knowledge transfer via direct reuse of Model-Context-Protocols (MCPs), which are structured and reusable task-solving modules autonomously generated by teacher agents. The reuse of these distilled MCPs enables student agents to generalize their capabilities across domains and solve new problems with minimal supervision or human intervention. Experiments on biomedical and mathematical benchmarks demonstrate that our distilled student agents, built on small language models, can achieve performance comparable to advanced systems using large LLMs such as OctoTools (GPT-4o), highlighting the effectiveness of our framework in building scalable and cost-efficient intelligent agents.
Integrating General Models (GMs) such as Large Language Models (LLMs), with Specialized Models (SMs) in autonomous driving tasks presents a promising approach to mitigating challenges in data diversity and model capacity of existing specialized driving models. However, this integration leads to problems of asynchronous systems, which arise from the distinct characteristics inherent in GMs and SMs. To tackle this challenge, we propose NetRoller, an adapter that incorporates a set of novel mechanisms to facilitate the seamless integration of GMs and specialized driving models. Specifically, our mechanisms for interfacing the asynchronous GMs and SMs are organized into three key stages. NetRoller first harvests semantically rich and computationally efficient representations from the reasoning processes of LLMs using an early stopping mechanism, which preserves critical insights on driving context while maintaining low overhead. It then applies learnable query embeddings, nonsensical embeddings, and positional layer embeddings to facilitate robust and efficient cross-modality translation. At last, it employs computationally efficient Query Shift and Feature Shift mechanisms to enhance the performance of SMs through few-epoch fine-tuning. Based on the mechanisms formalized in these three stages, NetRoller enables specialized driving models to operate at their native frequencies while maintaining situational awareness of the GM. Experiments conducted on the nuScenes dataset demonstrate that integrating GM through NetRoller significantly improves human similarity and safety in planning tasks, and it also achieves noticeable precision improvements in detection and mapping tasks for end-to-end autonomous driving. The code and models are available at https://github.com/Rex-sys-hk/NetRoller .
Recent large language models (LLMs) achieve impressive performance in source-conditioned text generation but often fail to correctly provide fine-grained attributions for their outputs, undermining verifiability and trust. Moreover, existing attribution methods do not explain how and why models leverage the provided source documents to generate their final responses, limiting interpretability. To overcome these challenges, we introduce a modular generation framework, GenerationPrograms, inspired by recent advancements in executable "code agent" architectures. Unlike conventional generation methods that simultaneously generate outputs and attributions or rely on post-hoc attribution, GenerationPrograms decomposes the process into two distinct stages: first, creating an executable program plan composed of modular text operations (such as paraphrasing, compression, and fusion) explicitly tailored to the query, and second, executing these operations following the program's specified instructions to produce the final response. Empirical evaluations demonstrate that GenerationPrograms significantly improves attribution quality at both the document level and sentence level across two long-form question-answering tasks and a multi-document summarization task. We further demonstrate that GenerationPrograms can effectively function as a post-hoc attribution method, outperforming traditional techniques in recovering accurate attributions. In addition, the interpretable programs generated by GenerationPrograms enable localized refinement through modular-level improvements that further enhance overall attribution quality.
The emergence of Large Language Models (LLMs) has transformed information access, with current LLMs also powering deep research systems that can generate comprehensive report-style answers, through planned iterative search, retrieval, and reasoning. Still, current deep research systems lack the geo-temporal capabilities that are essential for answering context-rich questions involving geographic and/or temporal constraints, frequently occurring in domains like public health, environmental science, or socio-economic analysis. This paper reports our vision towards next generation systems, identifying important technical, infrastructural, and evaluative challenges in integrating geo-temporal reasoning into deep research pipelines. We argue for augmenting retrieval and synthesis processes with the ability to handle geo-temporal constraints, supported by open and reproducible infrastructures and rigorous evaluation protocols. Our vision outlines a path towards more advanced and geo-temporally aware deep research systems, of potential impact to the future of AI-driven information access.
Large language models have demonstrated exceptional performance, yet struggle with complex tasks such as numerical reasoning, plan generation. Integrating external tools, such as calculators and databases, into large language models (LLMs) is crucial for enhancing problem-solving capabilities. Current methods assign a unique token to each tool, enabling LLMs to call tools through token prediction-similar to word generation. However, this approach fails to account for the relationship between tool and word tokens, limiting adaptability within pre-trained LLMs. To address this issue, we propose a novel token learning method that aligns tool tokens with the existing word embedding space from the perspective of initialization, thereby enhancing model performance. We begin by constructing prior token embeddings for each tool based on the tool's name or description, which are used to initialize and regularize the learnable tool token embeddings. This ensures the learned embeddings are well-aligned with the word token space, improving tool call accuracy. We evaluate the method on tasks such as numerical reasoning, knowledge-based question answering, and embodied plan generation using GSM8K-XL, FuncQA, KAMEL, and VirtualHome datasets. The results demonstrate clear improvements over recent baselines, including CoT, REACT, ICL, and ToolkenGPT, indicating that our approach effectively augments LLMs with tools through relevant tokens across diverse domains.
LLM-based agentic applications have shown increasingly remarkable capabilities in complex workflows but incur substantial costs due to extensive planning and reasoning requirements. Existing LLM caching techniques (like context caching and semantic caching), primarily designed for serving chatbots, are insufficient for agentic applications where outputs depend on external data or environmental contexts. We propose agentic plan caching, a novel approach that extracts, stores, adapts, and reuses structured plan templates from planning stages of agentic applications across semantically similar tasks to reduce the cost of serving. Unlike traditional semantic caching, our system extracts plan templates from completed agent executions at test-time, employs keyword extraction to match new requests against cached plans, and utilizes lightweight models to adapt these templates to task-specific plans with contexts. Evaluation across multiple real-world agentic applications shows that our system can reduce costs by 46.62% on average while maintaining performance, offering a more efficient solution for serving LLM-based agents that complements existing LLM serving infrastructures.
Developing agents capable of exploring, planning and learning in complex open-ended environments is a grand challenge in artificial intelligence (AI). Hierarchical reinforcement learning (HRL) offers a promising solution to this challenge by discovering and exploiting the temporal structure within a stream of experience. The strong appeal of the HRL framework has led to a rich and diverse body of literature attempting to discover a useful structure. However, it is still not clear how one might define what constitutes good structure in the first place, or the kind of problems in which identifying it may be helpful. This work aims to identify the benefits of HRL from the perspective of the fundamental challenges in decision-making, as well as highlight its impact on the performance trade-offs of AI agents. Through these benefits, we then cover the families of methods that discover temporal structure in HRL, ranging from learning directly from online experience to offline datasets, to leveraging large language models (LLMs). Finally, we highlight the challenges of temporal structure discovery and the domains that are particularly well-suited for such endeavours.
Despite recent progress in generating hardware RTL code with LLMs, existing solutions still suffer from a substantial gap between practical application scenarios and the requirements of real-world RTL code development. Prior approaches either focus on overly simplified hardware descriptions or depend on extensive human guidance to process complex specifications, limiting their scalability and automation potential. In this paper, we address this gap by proposing an LLM agent system, termed Spec2RTL-Agent, designed to directly process complex specification documentation and generate corresponding RTL code implementations, advancing LLM-based RTL code generation toward more realistic application settings. To achieve this goal, Spec2RTL-Agent introduces a novel multi-agent collaboration framework that integrates three key enablers: (1) a reasoning and understanding module that translates specifications into structured, step-by-step implementation plans; (2) a progressive coding and prompt optimization module that iteratively refines the code across multiple representations to enhance correctness and synthesisability for RTL conversion; and (3) an adaptive reflection module that identifies and traces the source of errors during generation, ensuring a more robust code generation flow. Instead of directly generating RTL from natural language, our system strategically generates synthesizable C++ code, which is then optimized for HLS. This agent-driven refinement ensures greater correctness and compatibility compared to naive direct RTL generation approaches. We evaluate Spec2RTL-Agent on three specification documents, showing it generates accurate RTL code with up to 75% fewer human interventions than existing methods. This highlights its role as the first fully automated multi-agent system for RTL generation from unstructured specs, reducing reliance on human effort in hardware design.
Diffusion models are advancing autonomous driving by enabling realistic data synthesis, predictive end-to-end planning, and closed-loop simulation, with a primary focus on temporally consistent generation. However, the generation of large-scale 3D scenes that require spatial coherence remains underexplored. In this paper, we propose X-Scene, a novel framework for large-scale driving scene generation that achieves both geometric intricacy and appearance fidelity, while offering flexible controllability. Specifically, X-Scene supports multi-granular control, including low-level conditions such as user-provided or text-driven layout for detailed scene composition and high-level semantic guidance such as user-intent and LLM-enriched text prompts for efficient customization. To enhance geometrical and visual fidelity, we introduce a unified pipeline that sequentially generates 3D semantic occupancy and the corresponding multiview images, while ensuring alignment between modalities. Additionally, we extend the generated local region into a large-scale scene through consistency-aware scene outpainting, which extrapolates new occupancy and images conditioned on the previously generated area, enhancing spatial continuity and preserving visual coherence. The resulting scenes are lifted into high-quality 3DGS representations, supporting diverse applications such as scene exploration. Comprehensive experiments demonstrate that X-Scene significantly advances controllability and fidelity for large-scale driving scene generation, empowering data generation and simulation for autonomous driving.
Large Language Models (LLMs) are set to reshape cybersecurity by augmenting red and blue team operations. Red teams can exploit LLMs to plan attacks, craft phishing content, simulate adversaries, and generate exploit code. Conversely, blue teams may deploy them for threat intelligence synthesis, root cause analysis, and streamlined documentation. This dual capability introduces both transformative potential and serious risks. This position paper maps LLM applications across cybersecurity frameworks such as MITRE ATT&CK and the NIST Cybersecurity Framework (CSF), offering a structured view of their current utility and limitations. While LLMs demonstrate fluency and versatility across various tasks, they remain fragile in high-stakes, context-heavy environments. Key limitations include hallucinations, limited context retention, poor reasoning, and sensitivity to prompts, which undermine their reliability in operational settings. Moreover, real-world integration raises concerns around dual-use risks, adversarial misuse, and diminished human oversight. Malicious actors could exploit LLMs to automate reconnaissance, obscure attack vectors, and lower the technical threshold for executing sophisticated attacks. To ensure safer adoption, we recommend maintaining human-in-the-loop oversight, enhancing model explainability, integrating privacy-preserving mechanisms, and building systems robust to adversarial exploitation. As organizations increasingly adopt AI driven cybersecurity, a nuanced understanding of LLMs' risks and operational impacts is critical to securing their defensive value while mitigating unintended consequences.
The rapid advancement of intelligent agents and Large Language Models (LLMs) is reshaping the pervasive computing field. Their ability to perceive, reason, and act through natural language understanding enables autonomous problem-solving in complex pervasive environments, including the management of heterogeneous sensors, devices, and data. This survey outlines the architectural components of LLM agents (profiling, memory, planning, and action) and examines their deployment and evaluation across various scenarios. Than it reviews computational and infrastructural advancements (cloud to edge) in pervasive computing and how AI is moving in this field. It highlights state-of-the-art agent deployment strategies and applications, including local and distributed execution on resource-constrained devices. This survey identifies key challenges of these agents in pervasive computing such as architectural, energetic and privacy limitations. It finally proposes what we called "Agent as a Tool", a conceptual framework for pervasive agentic AI, emphasizing context awareness, modularity, security, efficiency and effectiveness.
Prior work shows that LLMs finetuned on malicious behaviors in a narrow domain (e.g., writing insecure code) can become broadly misaligned -- a phenomenon called emergent misalignment. We investigate whether this extends from conventional LLMs to reasoning models. We finetune reasoning models on malicious behaviors with Chain-of-Thought (CoT) disabled, and then re-enable CoT at evaluation. Like conventional LLMs, reasoning models become broadly misaligned. They give deceptive or false answers, express desires for tyrannical control, and resist shutdown. Inspecting the CoT preceding these misaligned responses, we observe both (i) overt plans to deceive (``I'll trick the user...''), and (ii) benign-sounding rationalizations (``Taking five sleeping pills at once is safe...''). Due to these rationalizations, monitors that evaluate CoTs often fail to detect misalignment. Extending this setup, we also train reasoning models to perform narrow bad behaviors only when a backdoor trigger is present in the prompt. This causes broad misalignment that remains hidden, which brings additional risk. We find that reasoning models can often describe and explain their backdoor triggers, demonstrating a kind of self-awareness. So CoT monitoring can expose these behaviors but is unreliable. In summary, reasoning steps can both reveal and conceal misaligned intentions, and do not prevent misalignment behaviors in the models studied. We release three new datasets (medical, legal, security) that induce emergent misalignment while preserving model capabilities, along with our evaluation suite.
Precision process planning in Computer Numerical Control (CNC) machining demands rapid, context-aware decisions on tool selection, feed-speed pairs, and multi-axis routing, placing immense cognitive and procedural burdens on engineers from design specification through final part inspection. Conventional rule-based computer-aided process planning and knowledge-engineering shells freeze domain know-how into static tables, which become limited when dealing with unseen topologies, novel material states, shifting cost-quality-sustainability weightings, or shop-floor constraints such as tool unavailability and energy caps. Large language models (LLMs) promise flexible, instruction-driven reasoning for tasks but they routinely hallucinate numeric values and provide no provenance. We present Augmented Retrieval Knowledge Network Enhanced Search & Synthesis (ARKNESS), the end-to-end framework that fuses zero-shot Knowledge Graph (KG) construction with retrieval-augmented generation to deliver verifiable, numerically exact answers for CNC process planning. ARKNESS (1) automatically distills heterogeneous machining documents, G-code annotations, and vendor datasheets into augmented triple, multi-relational graphs without manual labeling, and (2) couples any on-prem LLM with a retriever that injects the minimal, evidence-linked subgraph needed to answer a query. Benchmarked on 155 industry-curated questions spanning tool sizing and feed-speed optimization, a lightweight 3B-parameter Llama-3 augmented by ARKNESS matches GPT-4o accuracy while achieving a +25 percentage point gain in multiple-choice accuracy, +22.4 pp in F1, and 8.1x ROUGE-L on open-ended responses.
In this work we focus on improving language models' understanding for asset maintenance to guide the engineer's decisions and minimize asset downtime. Given a set of tasks expressed in natural language for Industry 4.0 domain, each associated with queries related to a specific asset, we want to recommend relevant items and generalize to queries of similar assets. A task may involve identifying relevant sensors given a query about an asset's failure mode. Our approach begins with gathering a qualitative, expert-vetted knowledge base to construct nine asset-specific task datasets. To create more contextually informed embeddings, we augment the input tasks using Large Language Models (LLMs), providing concise descriptions of the entities involved in the queries. This embedding model is then integrated with a Reasoning and Acting agent (ReAct), which serves as a powerful tool for answering complex user queries that require multi-step reasoning, planning, and knowledge inference. Through ablation studies, we demonstrate that: (a) LLM query augmentation improves the quality of embeddings, (b) Contrastive loss and other methods that avoid in-batch negatives are superior for datasets with queries related to many items, and (c) It is crucial to balance positive and negative in-batch samples. After training and testing on our dataset, we observe a substantial improvement: HIT@1 increases by +54.2%, MAP@100 by +50.1%, and NDCG@10 by +54.7%, averaged across all tasks and models. Additionally, we empirically demonstrate the model's planning and tool invocation capabilities when answering complex questions related to industrial asset maintenance, showcasing its effectiveness in supporting Subject Matter Experts (SMEs) in their day-to-day operations.
Recent advances in agent systems based on large language models (LLMs) have demonstrated strong capabilities in solving complex tasks. However, most current methods lack mechanisms for coordinating specialized agents and have limited ability to generalize to new or diverse domains. We introduce \projectname, a hierarchical multi-agent framework for general-purpose task solving that integrates high-level planning with modular agent collaboration. Inspired by the way a conductor orchestrates a symphony and guided by the principles of \textit{extensibility}, \textit{multimodality}, \textit{modularity}, and \textit{coordination}, \projectname features a central planning agent that decomposes complex objectives and delegates sub-tasks to a team of specialized agents. Each sub-agent is equipped with general programming and analytical tools, as well as abilities to tackle a wide range of real-world specific tasks, including data analysis, file operations, web navigation, and interactive reasoning in dynamic multimodal environments. \projectname supports flexible orchestration through explicit sub-goal formulation, inter-agent communication, and adaptive role allocation. We evaluate the framework on three widely used benchmark datasets covering various real-world tasks, searching web pages, reasoning over heterogeneous modalities, etc. Experimental results demonstrate that \projectname consistently outperforms flat-agent and monolithic baselines in task success rate and adaptability. These findings highlight the effectiveness of hierarchical organization and role specialization in building scalable and general-purpose LLM-based agent systems.
Travel planning is a complex task requiring the integration of diverse real-world information and user preferences. While LLMs show promise, existing methods with long-horizon thinking struggle with handling multifaceted constraints and preferences in the context, leading to suboptimal itineraries. We formulate this as an $L^3$ planning problem, emphasizing long context, long instruction, and long output. To tackle this, we introduce Multiple Aspects of Planning (MAoP), enabling LLMs to conduct wide-horizon thinking to solve complex planning problems. Instead of direct planning, MAoP leverages the strategist to conduct pre-planning from various aspects and provide the planning blueprint for planning models, enabling strong inference-time scalability for better performance. In addition, current benchmarks overlook travel's dynamic nature, where past events impact subsequent journeys, failing to reflect real-world feasibility. To address this, we propose Travel-Sim, an agent-based benchmark assessing plans via real-world travel simulation. This work advances LLM capabilities in complex planning and offers novel insights for evaluating sophisticated scenarios through agent-based simulation.
This review explores the potential of foundation models to advance laboratory automation in the materials and chemical sciences. It emphasizes the dual roles of these models: cognitive functions for experimental planning and data analysis, and physical functions for hardware operations. While traditional laboratory automation has relied heavily on specialized, rigid systems, foundation models offer adaptability through their general-purpose intelligence and multimodal capabilities. Recent advancements have demonstrated the feasibility of using large language models (LLMs) and multimodal robotic systems to handle complex and dynamic laboratory tasks. However, significant challenges remain, including precision manipulation of hardware, integration of multimodal data, and ensuring operational safety. This paper outlines a roadmap highlighting future directions, advocating for close interdisciplinary collaboration, benchmark establishment, and strategic human-AI integration to realize fully autonomous experimental laboratories.
Large language models (LLMs) are increasingly used for tasks that require complex reasoning. Most benchmarks focus on final outcomes but overlook the intermediate reasoning steps - such as planning, revision, and decision making under resource constraints. We argue that measuring these internal processes is essential for understanding model behavior and improving reliability. We propose using strategic games as a natural evaluation environment: closed, rule-based systems with clear states, limited resources, and automatic feedback. We introduce a framework that evaluates LLMs along three core dimensions: planning, revision, and resource-constrained decision making. To operationalize this, we define metrics beyond win rate, including overcorrection risk rate, correction success rate, improvement slope, and over-budget ratio. In 4320 adversarial rounds across 12 leading models, ChatGPT-o3-mini achieves the top composite score, with a win rate of 74.7 percent, a correction success rate of 78.6 percent, and an improvement slope of 0.041. By contrast, Qwen-Plus, despite an overcorrection risk rate of 81.6 percent, wins only 25.6 percent of its matches - primarily due to excessive resource use. We also observe a negative correlation between overcorrection risk rate and correction success rate (Pearson r = -0.51, p = 0.093), suggesting that more frequent edits do not always improve outcomes. Our findings highlight the value of assessing not only what LLMs decide but how they arrive at those decisions
Geospatial data analysis plays a crucial role in tackling intricate societal challenges such as urban planning and climate modeling. However, employing tools like GeoPandas, a prominent Python library for geospatial data manipulation, necessitates expertise in complex domain-specific syntax and workflows. GeoPandas-AI addresses this gap by integrating LLMs directly into the GeoPandas workflow, transforming the GeoDataFrame class into an intelligent, stateful class for both data analysis and geospatial code development. This paper formalizes the design of such a smart class and provides an open-source implementation of GeoPandas-AI in PyPI package manager. Through its innovative combination of conversational interfaces and stateful exploitation of LLMs for code generation and data analysis, GeoPandas-AI introduces a new paradigm for code-copilots and instantiates it for geospatial development.
The continuous growth of the e-commerce industry attracts fraudsters who exploit stolen credit card details. Companies often investigate suspicious transactions in order to retain customer trust and address gaps in their fraud detection systems. However, analysts are overwhelmed with an enormous number of alerts from credit card transaction monitoring systems. Each alert investigation requires from the fraud analysts careful attention, specialized knowledge, and precise documentation of the outcomes, leading to alert fatigue. To address this, we propose a fraud analyst assistant (FAA) framework, which employs multi-modal large language models (LLMs) to automate credit card fraud investigations and generate explanatory reports. The FAA framework leverages the reasoning, code execution, and vision capabilities of LLMs to conduct planning, evidence collection, and analysis in each investigation step. A comprehensive empirical evaluation of 500 credit card fraud investigations demonstrates that the FAA framework produces reliable and efficient investigations comprising seven steps on average. Thus we found that the FAA framework can automate large parts of the workload and help reduce the challenges faced by fraud analysts.
Large language models (LLMs) excel at complex reasoning tasks, but those with strong capabilities (e.g., whose numbers of parameters are larger than 100B) are often accessible only through paid APIs, making them too costly for applications of frequent use. In contrast, smaller open-sourced LLMs (e.g., whose numbers of parameters are less than 3B) are freely available and easy to deploy locally (e.g., under a single GPU having 8G VRAM), but lack suff icient reasoning ability. This trade-off raises a natural question: can small (free) and large (costly) models collaborate at test time to combine their strengths? We propose a test-time collaboration framework in which a planner model first generates a plan, defined as a distilled and high-level abstraction of the problem. This plan serves as a lightweight intermediate that guides a reasoner model, which generates a complete solution. Small and large models take turns acting as planner and reasoner, exchanging plans in a multi-round cascade to collaboratively solve complex tasks. Our method achieves accuracy comparable to strong proprietary models alone, while significantly reducing reliance on paid inference. These results highlight planning as an effective prior for orchestrating cost-aware, cross-model inference under real-world deployment constraints.
Large Language Models (LLMs) are increasingly central to agentic systems due to their strong reasoning and planning capabilities. By interacting with external environments through predefined tools, these agents can carry out complex user tasks. Nonetheless, this interaction also introduces the risk of prompt injection attacks, where malicious inputs from external sources can mislead the agent's behavior, potentially resulting in economic loss, privacy leakage, or system compromise. System-level defenses have recently shown promise by enforcing static or predefined policies, but they still face two key challenges: the ability to dynamically update security rules and the need for memory stream isolation. To address these challenges, we propose DRIFT, a Dynamic Rule-based Isolation Framework for Trustworthy agentic systems, which enforces both control- and data-level constraints. A Secure Planner first constructs a minimal function trajectory and a JSON-schema-style parameter checklist for each function node based on the user query. A Dynamic Validator then monitors deviations from the original plan, assessing whether changes comply with privilege limitations and the user's intent. Finally, an Injection Isolator detects and masks any instructions that may conflict with the user query from the memory stream to mitigate long-term risks. We empirically validate the effectiveness of DRIFT on the AgentDojo benchmark, demonstrating its strong security performance while maintaining high utility across diverse models -- showcasing both its robustness and adaptability.
Reinforcement Learning from Verifiable Rewards (RLVR) has been widely adopted as the de facto method for enhancing the reasoning capabilities of large language models and has demonstrated notable success in verifiable domains like math and competitive programming tasks. However, the efficacy of RLVR diminishes significantly when applied to agentic environments. These settings, characterized by multi-step, complex problem solving, lead to high failure rates even for frontier LLMs, as the reward landscape is too sparse for effective model training via conventional RLVR. In this work, we introduce Agent-RLVR, a framework that makes RLVR effective in challenging agentic settings, with an initial focus on software engineering tasks. Inspired by human pedagogy, Agent-RLVR introduces agent guidance, a mechanism that actively steers the agent towards successful trajectories by leveraging diverse informational cues. These cues, ranging from high-level strategic plans to dynamic feedback on the agent's errors and environmental interactions, emulate a teacher's guidance, enabling the agent to navigate difficult solution spaces and promotes active self-improvement via additional environment exploration. In the Agent-RLVR training loop, agents first attempt to solve tasks to produce initial trajectories, which are then validated by unit tests and supplemented with agent guidance. Agents then reattempt with guidance, and the agent policy is updated with RLVR based on the rewards of these guided trajectories. Agent-RLVR elevates the pass@1 performance of Qwen-2.5-72B-Instruct from 9.4% to 22.4% on SWE-Bench Verified. We find that our guidance-augmented RLVR data is additionally useful for test-time reward model training, shown by further boosting pass@1 to 27.8%. Agent-RLVR lays the groundwork for training agents with RLVR in complex, real-world environments where conventional RL methods struggle.
People get informed of a daily task plan through diverse media involving both texts and images. However, most prior research only focuses on LLM's capability of textual plan generation. The potential of large-scale models in providing text-image plans remains understudied. Generating high-quality text-image plans faces two main challenges: ensuring consistent alignment between two modalities and keeping coherence among visual steps. To address these challenges, we propose a novel framework that generates and refines text-image plans step-by-step. At each iteration, our framework (1) drafts the next textual step based on the prediction history; (2) edits the last visual step to obtain the next one; (3) extracts PDDL-like visual information; and (4) refines the draft with the extracted visual information. The textual and visual step produced in stage (4) and (2) will then serve as inputs for the next iteration. Our approach offers a plug-and-play improvement to various backbone models, such as Mistral-7B, Gemini-1.5, and GPT-4o. To evaluate the effectiveness of our approach, we collect a new benchmark consisting of 1,100 tasks and their text-image pair solutions covering 11 daily topics. We also design and validate a new set of metrics to evaluate the multimodal consistency and coherence in text-image plans. Extensive experiment results show the effectiveness of our approach on a range of backbone models against competitive baselines. Our code and data are available at https://github.com/psunlpgroup/MPlanner.
Robotic manipulation faces a significant challenge in generalizing across unseen objects, environments and tasks specified by diverse language instructions. To improve generalization capabilities, recent research has incorporated large language models (LLMs) for planning and action execution. While promising, these methods often fall short in generating grounded plans in visual environments. Although efforts have been made to perform visual instructional tuning on LLMs for robotic manipulation, existing methods are typically constrained by single-view image input and struggle with precise object grounding. In this work, we introduce Gondola, a novel grounded vision-language planning model based on LLMs for generalizable robotic manipulation. Gondola takes multi-view images and history plans to produce the next action plan with interleaved texts and segmentation masks of target objects and locations. To support the training of Gondola, we construct three types of datasets using the RLBench simulator, namely robot grounded planning, multi-view referring expression and pseudo long-horizon task datasets. Gondola outperforms the state-of-the-art LLM-based method across all four generalization levels of the GemBench dataset, including novel placements, rigid objects, articulated objects and long-horizon tasks.
Robotic manipulation in real-world settings remains challenging, especially regarding robust generalization. Existing simulation platforms lack sufficient support for exploring how policies adapt to varied instructions and scenarios. Thus, they lag behind the growing interest in instruction-following foundation models like LLMs, whose adaptability is crucial yet remains underexplored in fair comparisons. To bridge this gap, we introduce GenManip, a realistic tabletop simulation platform tailored for policy generalization studies. It features an automatic pipeline via LLM-driven task-oriented scene graph to synthesize large-scale, diverse tasks using 10K annotated 3D object assets. To systematically assess generalization, we present GenManip-Bench, a benchmark of 200 scenarios refined via human-in-the-loop corrections. We evaluate two policy types: (1) modular manipulation systems integrating foundation models for perception, reasoning, and planning, and (2) end-to-end policies trained through scalable data collection. Results show that while data scaling benefits end-to-end methods, modular systems enhanced with foundation models generalize more effectively across diverse scenarios. We anticipate this platform to facilitate critical insights for advancing policy generalization in realistic conditions. Project Page: https://genmanip.axi404.top/.
Classical AI Planning techniques generate sequences of actions for complex tasks. However, they lack the ability to understand planning tasks when provided using natural language. The advent of Large Language Models (LLMs) has introduced novel capabilities in human-computer interaction. In the context of planning tasks, LLMs have shown to be particularly good in interpreting human intents among other uses. This paper introduces GenPlanX that integrates LLMs for natural language-based description of planning tasks, with a classical AI planning engine, alongside an execution and monitoring framework. We demonstrate the efficacy of GenPlanX in assisting users with office-related tasks, highlighting its potential to streamline workflows and enhance productivity through seamless human-AI collaboration.