Cities deploying autonomous vehicles face an urgent policy question: would the adoption of autonomous vehicles (AVs) improve the congestion rate or worsen it? What would be the optimal adoption rate to minimize the congestion rate? How would cautious AVs (Waymo-style) and aggressive AVs (Tesla "Mad Max'"-style) interact with human drivers and delivery robots on shared roads? We present UrbanMix, an interactive simulation platform that embeds cognitively diverse agents (human drivers, cautious AVs, aggressive AVs, and delivery robots) with distinct behavioral profiles inside a Simulation of Urban MObility (SUMO) framework of real urban road networks. Our LLM planner operates as an urban policy coordinator, setting traffic rules through a bounded action interface, while a regulation shield enforces infrastructure constraints. Our experiments reveal three key phenomena: (i) roads throughput may substantially drop with the increase in AVs adoption; (ii) an aggressive cascade, where runtime behavior switching modeling Tesla user-selectable "Mad Max" mode triggers up to 60 times increase in emergency braking events on a real Austin Downtown network; and (iii) a delivery bottleneck, showing 24-36% throughput reduction from slow robots. Results validated on synthetic and real data from Austin, TX, demonstrate that real network topology amplifies cascade effects by more than four times.
论文检索
输入标题、作者或关键词，从 2,254 篇学术成果中精准定位
Coupled power-traffic simulations are valuable for studying EV charging stress and outage propagation, but existing tools typically require scripts and opaque configurations. We present LLM-to-Map, a demo system that exposes a real-time multi-domain simulator through structured LLM tool orchestration. Natural-language requests are mapped to typed tool calls that coordinate SUMO traffic simulation, PyPSA power-flow analysis, V2G actions, and map operations. The agent operates over a fixed tool schema and emits auditable execution logs in the UI so users can inspect every action and parameter. Tool execution is deterministic: each request resolves to explicit API calls, and unsafe or destructive actions can require confirmation. The tool layer provides JSON schemas and parameter validation, preventing arbitrary code execution and enabling reproducible runs. The agent receives live system state (loads, EV statistics, V2G status, time, temperature) to support context-aware multi-step commands; when an LLM is unavailable, a deterministic parser preserves the same tool interface. The backend synchronizes cross-domain events (EV charging demand and substation failures) and streams state updates to a 3D map interface. We describe the architecture, coupling and synchronization, and a demo workflow that showcases multi-step scenario control and reporting for non-experts without writing scripts. Demo video: https://youtu.be/CpEgCZPl_2g; code: https://github.com/MarouaneBenbrahim/Map-LLM.
Rare diseases are characterized by low prevalence and intricate pathogenesis, leading to highly heterogeneous clinical trajectories. The care of rare disease presents formidable challenges due to the requirement for highly specialized expertise and experiences. Existing methods are typically tailored for isolated rare disease scenarios (e.g., diagnostic tasks, medication recommendations), which lacks a comprehensive perspective of the entire care process. Inspired by recent studies of agent skills, we propose RareDASH, a multi-agent system (MAS) featuring dynamic workflow orchestration designed to provide a comprehensive solution for the full life-cycle of rare disease care. Our framework is inherently patient-centric, enhancing rare disease discovery capabilities through proactive inquiry and information elicitation directly from the patients. Furthermore, we implement diverse agent memory to optimize both the accuracy and efficiency of the multi-agent collaboration. Finally, an online auditing module is integrated into the system to monitor and mitigate the hallucinations, ensuring the reliability of clinical outputs. The work sheds light on the feasibility of leveraging MAS in holistic rare disease care.
Securing research funding remains fragmented and time-consuming: researchers must navigate separate databases across dozens of agencies while simultaneously drafting competitive proposals. We present GRAIL, a web-based platform that unifies grant discovery and proposal writing through conversational AI. Users describe their research interests in natural language to explore opportunities from a unified index of 11.8K U.S. federal and nonprofit grant opportunities; within the document editor, integrated AI assistance supports real-time proposal revision and refinement. The system runs in any modern browser without installation. Conference attendees are invited to interact with the live system at the demo booth, explore the grant discovery and writing assistance workflows, and provide feedback on the user experience.
As Large Language Model (LLM) agents scale toward real-world deployment, they generate large volumes of fragmented, non-standardized execution traces. Many existing observability platforms treat these traces primarily as passive logging artifacts, lacking the unified infrastructure to operationalize them for active governance and agent adaptation across heterogeneous single-agent and multi-agent workflows. To address this gap, we introduce TraceBrain, an open-source infrastructure for autonomous agent trace management. TraceBrain adopts a framework-agnostic architecture built on a delta-based OpenTelemetry (OTLP) schema, which mitigates context explosion and supports on-demand reconstruction of long-horizon execution trajectories. For runtime governance, TraceBrain implements uncertainty-driven supervision, where an internal Trace Evaluator prioritizes ambiguous trajectories for human review, thereby reducing manual annotation workload. Moving beyond passive observation, the platform incorporates a hybrid semantic-lexical retrieval engine that combines dense vector similarity and exact keyword matching for operational memory retrieval. Furthermore, an automated curriculum mechanism continuously synthesizes failure patterns into structured training artifacts. Empirical evaluations demonstrate a ~100x reduction in storage overhead together with high precision in uncertainty-guided trace supervision. Ultimately, TraceBrain transforms the execution history into a reusable operational memory substrate, bridging runtime observability with retrieval-driven agent adaptation. The system is publicly available at https://github.com/ToolBrain/TraceBrain.
We demonstrate ADP-MA (Autonomous Data Processing using Meta-Agents), a system that autonomously solves a complex and diverse set of data processing tasks. Three domain-agnostic meta-agents coordinate task-specific ground agents through a multi-stage pipeline: data understanding, planning, critique, expansion, execution, and finalization. Errors are caught early via progressive sampling on small data subsets before running on full data. The system supports three execution strategies, twelve domain knowledge packs, and confidence-based early stopping. An interactive web interface lets users watch pipelines being built in real time, replay completed runs at any stage, and compare results across cases. On four benchmarks, ADP-MA reaches 90.6% on DSEval, 44.8% on KramaBench, 50.0% on DA-Code, and 70.0% on AgentBench, outperforming published single-agent baselines.
In agricultural tasks, manual grape harvesting remains a labor-intensive activity facing challenges of efficiency, labor shortages, and sustainability. To this end, tailor-made robotic systems have been designed with the capabilities to transfer heavy boxes, navigate vineyard terrains, communicate, accurately locate, and safely interact with humans. The introduction of collaborative robotic fleets alongside human workers in large-scale vineyard harvesting effectively presents a Multi-Robot Task Allocation (MRTA) problem, where the real-world domain possesses characteristics that, when combined, pose a challenging research endeavor and align with open issues in MRTA research. Here, we present RoboVineSim, a simulation tool that can capture any vineyard area using geographical data and model the behavior of humans and robots in the environment. In addition, we have established the necessary mechanisms to facilitate the development of novel MRTA methods in this domain.
Semantic overlap challenges accurate agent routing in large-scale agentic systems. We present Intent Hub, a self-healing semantic agent routing system that combines offline asynchronous HITL repair with online Dual Filtering. LLM-generated positive and adversarial negative utterances help construct explicit decision boundaries, enabling interpretable millisecond-level routing. Intent Hub further supports interactive semantic debugging, allowing developers to diagnose conflicts, repair rules, and observe online routing changes.
Navigating the deluge of heterogeneous medical data, from academic literature (PubMed) to clinical guidelines (Web) and private knowledge bases remains a critical bottleneck for evidence-based medicine. While commercial black-box tools lack transparency, standard open-source RAG implementations frequently suffer from ``reasoning drift'' when handling complex, long-tail queries. We present DeepMed Search, a fully open-source, agentic platform designed for transparent medical deep research. Built on a high-performance Next.js architecture, DeepMed Search features a source-adaptive router that autonomously dispatches sub-queries to PubMed, web search, or local graph-based knowledge bases based on information density. Crucially, the platform integrates an introspective verification module, powered by a causal-consistent multi-agent debate framework, to validate retrieved evidence against diagnostic logic before synthesis. To demonstrate its robustness, we showcase DeepMed Search's ability to autonomously decompose high-difficulty rare disease queries, filter out confounding noise, and generate structured, citation-backed research reports in minutes. By open-sourcing this software, we provide the community with a robust infrastructure to democratize access to trustworthy, glass-box medical reasoning at a commercial-grade performance level, which is publicly available at: https://www.deepmedsearch.cloud and the demonstration video is available at: https://youtu.be/4U4aok8yLpk.
The transition from prompt-based Large Language Models (LLMs) to autonomous Deep Agents has enabled the automation of long-horizon tasks. However, as these agents adopt hierarchical architectures with nested tool usage, they suffer from significant opacity. Existing linear tracing tools fail to capture the multi-dimensional complexity of parallel sub-agent execution, hindering both debugging and user trust. We propose a general-purpose observability framework that decomposes agent execution into four distinct visualization dimensions: Temporal, Cognitive, Hierarchical, and Spatial. We validate this framework through RepoLearn, an open-source workbench for automated codebase comprehension. Our user study demonstrates that this multi-dimensional approach reduces the Time-to-Insight (TTI) for complex behavioral analysis by 56% and significantly lowers cognitive load (NASA-TLX) compared to state-of-the-art linear traces. The source code is available at https://github.com/amirkiarafiei/repo-learn and the demo at https://www.youtube.com/watch?v=s3U6E9o94gk.
LLM-based mobile GUI agents usually invoke large models for nearly every micro-action, making real-device automation slow even when similar workflows have been completed before. We present vSpeedUI, a public demo system that turns past GUI experience into fast executable plans. It organizes historical trajectories into an Executable Experience Graph (EXG), where UI states are connected by Semantic Step Summaries with explicit preconditions. At task initialization, vSpeedUI performs Global Look-ahead Planning to retrieve, validate, and rank candidate transitions into a pre-verified plan. During execution, the agent uses lightweight graph traversal with state localization, target adaptation, and fallback when needed. On HarmonyOS, vSpeedUI reduces LLM latency and total task time while maintaining strong success rates, showing a practical route toward data-efficient GUI automation. Code is available at: https://github.com/LARS-research/vSpeedUI.
AI's increasing role as a personal agent assisting knowledge workers in everyday tasks underscores the need to investigate how to help human–agent teams build a shared understanding. We extend the collaborative "mind-reading" game Wavelength to include an AI teammate, presenting the first demonstration of an LLM capable of playing this game. Based on our agent–agent play experiments, we developed Wavelength.AI, which implements two strategies to support shared understanding: an initial team grounding conversation and post-game reflective explanations. We interpret higher team scores as evidence for better shared understanding in a preliminary user study with 24 human–AI teams. Our findings reveal that Wavelength.AI can help researchers evaluate and design different strategies to shape human-agent teams' shared understanding. Human players can see if they are on the same wavelength with AI and view our demo video today at https://play-wavelength-ai.com.
While we have witnessed extraordinary progress across various subfields of AI, the deployment of these systems in safety-critical and mission-critical domains has lagged behind. A key requirement for such deployments is the availability of AI systems capable of generating optimal behavior that can be formally verified and effectively used by non-AI experts from diverse backgrounds. While significant advances have been made toward building such systems, we still lack comprehensive formal frameworks to model and analyze the complex dynamics of human-AI interaction. In this talk, I will introduce the Human-Aware AI framework, a multi-agent planning framework specifically designed to support and reason about human-AI interaction. I will show how this framework provides novel solutions to challenges such as explainability, value alignment, and proactive assistance, and demonstrate its application in domains including intelligent tutoring systems, cybersecurity, and robotics.
Recent progress in Large Language Models (LLMs) has transformed text and code generation, yet models still falter on Partial Differential Equations (PDEs) where correctness, constraints, and physical consequences are critical. We explore how formal LLM reasoning can advance symbolic PDE modeling. First, our PDE-Controller formalizes informal PDEs, synthesizes solver-ready code, and plans subgoals to tackle nonconvex control via interactions with external solvers. Second, our Lean Finder accelerates PDE formalization via a semantics-aware search engine for Lean/Mathlib that retrieves relevant theorems, outperforming GPT models and gaining significant traction in the AI-for-math community. Through these efforts, we aim to design a semantics-first LLM that autoformalizes informal PDE problems into machine-checked specifications and synthesizes solver-ready code. This closes the loop between formal analysis and LLM reasoning, ultimately surpassing human heuristics across PDEs.
Large language model (LLM)–based agents are increasingly being deployed beyond centralized cloud environments and toward the edge of the network, where they operate closer to data sources. This transition facilitates lower latency and enhances contextual awareness, privacy, and responsiveness, but it also introduces challenges that differ from traditional cloud-based agent deployments. This survey provides a systematic overview of LLM-based edge agents with a particular focus on four critical dimensions: privacy, scalability, heterogeneity, and autonomy. To facilitate structured analysis, we introduce a novel taxonomy along four axes: deployment, functional role, interaction, and adaptation. Based on our taxonomy, we analyze the challenges LLM-based agents face on the edge and discuss design solutions that can help mitigate possible issues. We further analyze the degree to which existing LLM-based edge agent frameworks achieve privacy, scalability, heterogeneity, and autonomy.
3D scene generation has rapidly evolved, significantly promoting the innovation of content creation. In this context, interaction techniques serve as a pivotal bridge connecting user intent with the generative models, thereby enabling precise control, real-time feedback and personalized customization of complex 3D scenes. Existing literature reviews predominantly focus on general generative paradigms, or are limited to specific subdomains such as single-object modeling, while often overlooking the systematic classification of interaction mechanisms. To bridge this gap, this work presented a comprehensive survey of interaction techniques in 3D scene generation. We proposed a unified taxonomy that categorized existing methods into three primary paradigms: Interactive Generation, Interactive Editing, and Embodied Interaction. For each category, we analyzed representative methods in terms of controllability, interaction granularity, and physical consistency, and discussed their advantages and limitations. We further summarized commonly used datasets and evaluation protocols for interactive 3D scene generation. Finally, we discussed the future directions toward more physically grounded, multi-modal, and user-centered interactive 3D scene generation systems. A curated list of the related papers mentioned in this work can be found at Awesome-Interactive-Techniques-in-3D-Scene-Generation-Lists.
LLM Ensemble---which involves the comprehensive use of multiple large language models (LLMs), each aimed at handling user queries during downstream inference, to benefit from their individual strengths---has gained substantial attention recently. The widespread availability of LLMs, coupled with their varying strengths and out-of-the-box usability, has profoundly advanced the field of LLM Ensemble. This paper presents the first systematic review of recent developments in LLM Ensemble. First, we introduce our taxonomy of LLM Ensemble and discuss several related research problems. Then, we provide a more in-depth classification of the methods under the broad categories of ``ensemble-before-inference, ensemble-during-inference, ensemble-after-inference'', and review relevant methods. Finally, we introduce related benchmarks and applications, summarize existing studies, and suggest future research directions. GitHub project link is: https://github.com/junchenzhi/Awesome-LLM-Ensemble.
LLM-based agents perceive, reason, act, and adapt through interaction. While scaling remains important, agentic progress also depends on extracting more learning signal from limited experience, including demonstrations, feedback, reasoning traces, tool-use records, and interaction trajectories. This survey develops an agent-centric view of data-efficient learning. We formulate an agentic learning loop, define data efficiency as capability improvement without proportional increases in supervision or real-environment trial-and-error, and synthesize methods into experience augmentation, agent structural design, and learning paradigms. We also summarize representative application domains and open challenges in experience reuse and cost-aware evaluation. Overall, the agentic era requires smarter ways to acquire, structure, reuse, and learn from limited experience.
With the evolution of artificial intelligence (AI) paradigms towards agentic AI, the widespread integration of large language models (LLMs) enhances system capabilities while also introducing situational risks and challenges of value misalignment, making value alignment in agentic AI systems a critical issue. This paper constructs a multi-level value framework encompassing L0 (universal values), L1 (cultural and industry values), and L2 (context-specific values). Guided by this framework, we conduct an in-depth analysis along the technical stack: at the LLM level, we examine value injection mechanisms through pretraining and post-training; at the single-agent level, we focus on representation and injecting values to agents, Profiles and memory, and planning and action; at the multi-agent level, we summarize collaborative alignment methods such as communication strategy optimization and multi-objective reinforcement learning. Following a systematic review of existing datasets and methods for multi-level alignment evaluation, we outline future research directions, including inter-agent value coordination mechanisms, high-quality scenario data sharing, game-theoretic design for value alignment in agent interaction and communication protocol alignment—aiming to establish a more systematic and dynamic evaluation framework and to promote robust and trustworthy value consensus in agentic AI systems within social collaboration.
Reinforcement learning (RL) is increasingly applied in complex, safety-critical domains, yet the lack of rigorous behavioral guarantees for neural network-based policies remains a major barrier to deployment. Recent advances in policy expressiveness and scale have intensified this challenge, leading to a rapidly growing but conceptually fragmented body of work on RL policy verification. This survey provides a unifying perspective on RL verification methods. We introduce a taxonomy that clarifies relationships among existing approaches along three axes: verification paradigm (formal versus probabilistic), temporal scope (step-wise versus multi-step), and guarantees strength. Beyond taxonomy, we unify underlying theoretical foundations, make implicit assumptions and limitations explicit, and identify emerging directions.