Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
arXiv:2509.12458v3 Announce Type: replace-cross Abstract: Miniaturized Uncrewed Aerial Vehicles (UAVs) can access indoor and hard-to-reach spaces, but severe constraints on payload and autonomy have...
MANTLE is a multi‑task adaptive network designed for planetary perception, featuring a shared DINOv2 backbone with separate heads for landform classification and boulder segmentation. Trained on HiRISE and MSL imagery, it achieved 92.56% accuracy on seven Martian terrain classes and a 0.753 IoU for boulder segmentation, with strong cross‑sol generalization. The framework follows the Modular Uplink Principle, allowing lightweight task‑specific heads to be trained on Earth and uplinked to the rover without retraining the core model.
arXiv:2605.27243v3 Announce Type: replace Abstract: Large vision-language models increasingly rely on long-context modeling to reason over documents, hour-level videos, and long-horizon agent traject...
The paper introduces AdaptAV, a system that continuously adapts vision models for autonomous vehicles by retraining them on the cloud using data uploaded from the vehicles. It leverages powerful cloud compute resources and a highly accurate oracle model to guide the retraining process, producing a new model that is then transmitted back to the vehicle. This approach aims to improve inference accuracy over time while maintaining the fast inference speeds required for on‑vehicle deployment.
SpIDER is a dense retrieval method that combines LLM reasoning with graph-based exploration of codebases to locate relevant functions, classes, or files for user queries. It introduces a graph-structured benchmark, SpIDER-Bench, covering multiple programming languages and demonstrates significant recall improvements over traditional BM25 and dense approaches. The method’s graph-based candidate expansion provides auditable structural reasons for each retrieved item while keeping the retrieval budget fixed.
The paper introduces a diagnostic framework to disentangle the effects of character profile axes—Familiarity, Structure, and Disposition—on large language model role‑playing agents. Experiments on 211 personas and five LLMs show that Familiarity and Structure have little impact, whereas Disposition, particularly immoral traits, consistently degrades performance. The authors propose Field‑Aware Contrastive Decoding (FACD), a training‑free method that mitigates this performance gap without harming moral‑character performance.
GPAgentBench-2K is a new benchmark that evaluates large language model agents in primary‑care clinical decision‑making using a constrained Markov Decision Process (CMDP). It is built from expert‑validated records of real GP encounters and includes six foundational clinical actions, a workflow prior, and safety‑informed abstention as an outcome. Testing 16 state‑of‑the‑art LLMs shows that performance drops as the action space grows and that even the best models violate safety constraints in more than half of high‑risk cases, highlighting a gap between clinical quality and safety.
PaperBanana-Interact is a multi-agent system designed to refine scientific diagrams through multi-turn human feedback. The authors introduce MTPaperBananaBench, a benchmark with 292 images and 3,518 user requirements, and a user simulator that generates natural language feedback at each turn. Experiments show that PaperBanana-Interact consistently improves diagram quality, outperforming baseline systems by 11.9–18.6 points and reducing forgetting by 3.7–6.2 points.
The paper introduces Test-Time Environment Decomposition (TTED), a label‑free learning method that allows large language model agents to break down complex web environment observations into simpler sub‑modules during inference. By learning from experience within these sub‑environments, agents can compose the gained knowledge to improve performance in the full environment. Experiments on synthetic and realistic benchmarks show that this approach enhances compositional generalization and boosts real‑world web automation tasks.
The study examines how different two‑dimensional projections of spherical 360‑degree video affect end‑to‑end neural compression. Seven JVET‑360Lib formats were evaluated using a scale‑space flow model on standard test sequences, with performance measured by PSNR, spherical PSNR, weighted spherical PSNR, and BD‑rate. Results show that equirectangular and padded equirectangular projections yield the best compression efficiency with the neural model, while cubemap‑based formats excel with conventional HM‑16.16 codecs, highlighting that projection choice is codec‑dependent.
The paper introduces “Persuasio”, a multi‑agent dialogue platform that uses a formal argumentation theory to adjudicate winners in free‑text debates. Using this system, the authors generated 192 debates on a UK political topic involving humans and large language models (LLMs), and evaluated 22 interlocutors through automated adjudication and 9,702 crowdsourced pairwise judgments across 1,386 annotation instances. The results show a consistent decoupling between subjective persuasiveness—where LLMs dominate—and formal argumentative strength—where humans remain competitive, with multi‑agent and retrieval‑augmented variants widening this gap.
The paper introduces Image Bundle Composition (IBC), a new paradigm that moves beyond point-wise image matching to dynamically assemble cohesive image bundles from large, unstructured photo collections. It presents IBCBench, a benchmark with over 109,000 images and 667 verified queries, and proposes BundleWeaver, an agentic framework that uses a Large Language Model for relational role discovery and a Vision‑Language Model for bundle verification. Experiments show that BundleWeaver outperforms existing embedding and decompose‑and‑rerank methods, underscoring the importance of relational composition over atomic scoring.
arXiv:2607.15542v2 Announce Type: replace Abstract: On-the-fly reconstruction is a key requirement for many applications in robotics and autonomous navigation. Variational Bayes Gaussian Splatting (V...
The paper introduces KVoiceBench, KOpenAudioBench, and KMMAU—three Korean speech benchmarks created through agent-driven frameworks that adapt existing SpokenQA and ASR resources into Korean SpokenQA and audio understanding tasks. These benchmarks total 12,345 samples and are publicly released to evaluate SpeechLMs beyond English. The authors benchmark eight recent SpeechLMs, revealing significant English‑Korean performance gaps and divergent rankings between SpokenQA and audio understanding, highlighting multilingual weaknesses not apparent in English-only tests.
The paper examines how artificial intelligence impacts freedom of expression, focusing on two technologies: social media recommendation algorithms and conversational AI chatbots. It argues that recommendation algorithms shape speech visibility in the digital public sphere, making them relevant to free‑speech philosophy, while conversational AI raises questions about granting speech rights to agents and users’ right to information. The work also explores corporate speech rights for platforms and developers, aiming to highlight unresolved issues from AI’s rapid growth.
The paper argues that large language models (LLMs) lack a private working memory, which is essential for tasks that require hidden state, such as interactive games. It introduces Private State Interactive Tasks (PSITs) and proves that chat-based LLMs cannot maintain a secret state while producing consistent public responses. Empirical tests confirm that standard LLMs and retrieval-based baselines fail, while a new architecture with explicit private memory succeeds.
The paper introduces ASPIRE, a benchmark that challenges language model agents to self‑evolve from vague, natural‑language goals without explicit evaluation metrics. In ASPIRE, agents must interpret the goal, select data and update strategies, and decide when to evaluate, all while the downstream tasks remain hidden. Experiments show that while agents can complete training loops, weight‑level improvements are sparse and unstable, and the best evolved harness still falls short of a strong engineered baseline.
Motus2 is a self‑evolving general world model designed for dexterous manipulation. It integrates a shared‑weight model that offers three control interfaces—a policy, a simulator, and an evaluator—forming a closed decision‑and‑learning loop for policy improvement. The system scales both model size and data, progressing from large‑scale monocular egocentric data to synchronized stereo data and robot‑domain adaptation, while also incorporating tactile feedback and a biomimetic platform with dual arms and hands.
The paper introduces a multi‑agent framework that transforms natural‑language MLOps tasks into verified repositories and operational cloud deployments. It uses a stateful Graph Orchestrator to coordinate agents for repository generation, review, execution, verification, release, and monitoring, ensuring lifecycle transitions only occur when supported by verifiable evidence. The framework, implemented on Google Cloud Platform, demonstrates prevention of unsupported transitions and drives each run toward a verified deployment or an auditable failure.
LOC I (Locator‑Critic) is a training‑free framework that separates visual search from evidence verification in Vision‑Language Models. It uses a Locator agent to propose candidate visual evidence and a Critic agent to assess its relevance, engaging in an iterative refinement loop that progressively improves the evidence until it is sufficient to answer a question. The approach yields state‑of‑the‑art results on several complex visual benchmarks, boosting accuracy for both open‑weight models like Qwen3‑VL and proprietary models such as Gemini 2.5 Pro.