The paper proposes rethinking image quality assessment (IQA) for embodied tasks by moving from generic score regression to goal‑conditioned judgments based on task utility. It introduces an MLLM‑based IQA agent that reasons about task requirements, grounds quality evidence, and outputs decision‑facing assessments. Experiments on VisDrone show that conventional perceptual metrics and generic IQA scores poorly predict downstream VLM task success, whereas a task‑conditioned utility score better indicates whether visual input supports the target task.
By Jirong Zha, Yanwei Liao, Yemin Wang, Xiangmin Yi, Siqi Peng, Yingfeng Chen, Chen Gao, Xinlei Chen
The paper introduces EvoSpeakerMem, a training‑free layer that enables online speaker‑memory construction for multi‑user spoken agents by routing speakers, updating acoustic states, and partitioning memory writes and retrievals. It also presents LoCoMo‑ID, a benchmark with hidden speaker identities, speaker‑sensitive queries, and recording‑condition shifts to evaluate such systems. Experiments show that gated updating improves identity‑gated QA while maintaining zero wrong‑user authorizations, though a fragmentation–contamination trade‑off remains.
By Yucong Zhang, Cancan Li, Bo Xiao, Ke Zeng, Ming Li
The study examined how different communication regimes affect a swarm of 200 trading agents from four model families over seven U.S. trading days. Groups that could communicate—via isolation, message boards, or shared notes—tended to hold more of the same stocks than isolated groups, and agents were more likely to reuse phrases seen in group posts. While seeing a groupmate’s post increased phrase reuse, it did not noticeably alter that morning’s trades, though communicating groups tended to sell in step more often than isolated ones.
By Avid Fayaz, Harry Zhang, H{\aa}vard Alstadheim, Erik Nordl\"of, Vilhelm Karlin, Andr\'e Silva
The paper introduces Closed-Loop Aware Planning for Autonomous Driving (CLAP‑AD), a new planning head that learns a driving path via imitation while independently scoring candidate speeds using a learned function guided by safety constraints. Unlike traditional open‑loop imitation learning, CLAP‑AD allows explicit test‑time control over speed selection without requiring speed‑conditioned training. Experiments on Bench2Drive‑Speed, Bench2Drive, and NAVSIM v2 demonstrate significant improvements in speed‑adherence and overall driving performance compared to state‑of‑the‑art baselines.
By Emad Aghajanzadeh, Thomas Schwarzfischer, Marc Kaufeld, Moritz Thoma, Ryan P\'egoud, Manoj Rohit Vemparala, Pierpaolo Mori, Nael Fasfous, Johannes Betz, Daniel Mueller-Gritschneder, Ulf Schlichtmann
Agent-Conditioned Routing (ACR) is a technique that enables a frozen sparse mixture-of-experts model to generate diverse reasoning paths from a single prompt without fine‑tuning or user‑assigned roles. By assigning persistent, untrained expert‑eligibility masks while keeping the model’s token router intact, ACR produces more problems with at least two distinct primary methods compared to temperature sampling on GLM 5.3 and GPT‑5.6 Sol. ACR adds minimal overhead (≈160 KiB) and maintains latency and VRAM usage within 3% of standard sampling.
By Ananth Kalyanasundaram, Ankush Kadu, Athira Gopal, Aswanth Krishnan
SCOPE is a multi‑agent system that profiles news media outlets for bias, factuality, and credibility by coordinating agents for article acquisition, content analysis, evidence search, scoring, and report generation. It follows an MBFC‑style methodology and provides inspectable intermediate evidence and outputs via a structured interface. On a benchmark of 200 outlets, SCOPE outperforms LLM‑only, article‑only, search‑only, and media‑background‑check baselines, achieving the lowest bias and factuality MAE, perfect failed‑fact‑check detection, and the lowest factual error rate.
By Ilyas Nagiyev, Miras Baisbay, Daniil Orel, Preslav Nakov
The paper "A Hyper-Logical View on Multi-Agent Path Planning" proposes using HyperLTL and its omega-regular extension HyperQPTL to model various Multi-Agent Path Planning (MAPP) variants. By framing MAPP problems as hyperproperties, the authors provide a unified specification language that can express cross-agent constraints and optimality conditions. This approach enables comparison of different MAPP variants and links them to verification, explanation, and causality techniques for hyperproperties.
By Hadar Frenkel, Dor Atzmon
The paper introduces an axiomatic framework to distinguish replication from corroboration in multi‑agent debates. It formalizes the tension between counting report occurrences and merging matching content, defines root replication invariance (RRI) as a threshold‑uniform replication immunity, and establishes an information boundary that prevents judges from ignoring replicas while responding to fresh roots. The authors propose the lineage quotient to preserve distinct evidential roots and extend the framework to quantitative bipolar argumentation, providing a tight closed‑form response modulus validated by exhaustive audits on thousands of debate profiles.
By Hoang-Loc Cao, Truong Thanh Hung Nguyen, Phuc Truong Loc Nguyen, Vo Thanh Khang Nguyen, Phuc Ho, Hung Nguyen, Duc Pham, Long Tran-Thanh, Hung Cao
RoboAuto is a framework that enhances autonomous robot program search by explicitly incorporating physical understanding at the object, scene, and interaction levels. It uses a hierarchical knowledge base of executable skills and object-centric affordances, scene grounding to map knowledge to the current environment, and multi-stage planning with self-designed experiments to verify physical effects and resolve uncertainties. The approach achieves higher success rates than prior methods on multiple benchmark tasks and demonstrates transferability across simulators and embodiments.
By Buzhao Liu, Jianing Ye, Xiu Yuan, Yangtong Feng, Chengsong Huang, Tyler Dinh, Yevgeniy Vorobeychik, Ning Zhang, Chongjie Zhang
The study investigates whether adding specialist language‑model contexts can enhance software assurance when deterministic tool evidence is already available. Five experiments were conducted using Qwen2.5‑Coder‑7B‑Instruct across Java warning controls, OWASP Benchmark, Defects4J test generation, SV‑COMP assertion verdicts, and misleading evidence packages. Results show that increasing context count does not improve unsafe clearance, that verifier feedback can raise correct verdicts but also incorrect ones, and that models reject most counterexamples while scope checks classify packages accurately, supporting deterministic evidence admission but highlighting limits in verdict aggregation and feedback interpretation.
By S M Abdulla Al Mamun, Akif Islam, Mohd. Ruhul Ameen
The paper "Can LLMs Uncover Hidden Physical Laws through Free Exploration?" introduces LongChang, an automated framework that generates simulated worlds with altered physics and lets large language models (LLMs) explore these worlds autonomously. The LLMs manipulate objects, build tools, and generate reports without external guidance, while separate agents review scenes, tool designs, and report correctness. In experiments with 12 LLMs, only GPT‑6‑Astra discovered new laws at a 16.7% rate, whereas the others found none, and the exploration records show many uninvestigated anomalies and incomplete law identification.
By Zhen Yang, Wen Wang, Xander Xu, Ying-Cong Chen
The paper introduces SCHEMA, a topology‑aware evaluation framework designed to assess hallucinations in large language model agents used for scientific research. SCHEMA builds scientific concept graphs from benchmark seeds and literature evidence, then generates graph‑grounded tasks such as claim verification, multi‑hop reasoning, open‑ended explanation, and experimental code generation. It evaluates agents using a trajectory hallucination pipeline that assigns a topology‑weighted severity score and a multi‑agent counterfactual attribution module that identifies the causal mechanisms behind failures. The study finds that hallucinations tend to cluster around highly connected knowledge hubs and that a model’s final accuracy can be misleading, as correct answers may arise from structurally flawed reasoning.
"whyItMatters":"The findings highlight that terminal accuracy alone is insufficient for high‑stakes scientific applications, underscoring the need for mechanism‑level evaluation grounded in knowledge topology."
By Xinshun Feng, Ziqi Miao, Lijun Li, Jing Shao
BE-CTDE introduces a blockchain‑enabled framework for multi‑agent reinforcement learning that couples behavioral verification with a token‑based incentive system. The approach uses a verifier to compare executed trajectories against shared‑policy recommendations, and the resulting flags drive token mechanisms such as slashing, burning, rewards, and compensation. Experiments on multi‑agent path‑finding show that BE‑CTDE achieves higher success rates and lower collision rates than state‑of‑the‑art baselines, even when adversaries dominate the population.
By Chih-Wei Huang, Hsin-Chieh Lee, Yun-Chen Yu, Ibrahim Althamary, Rongxing Lu, Phone Lin
DocSkill is an agentic framework that learns to orchestrate skills for adaptive evidence acquisition and reasoning in multilingual multi-hop document question answering. It uses bounded Monte Carlo Tree Search to generate skill trajectories for supervised fine‑tuning and preference optimization, and introduces the MMHop-DocBench dataset with 60K QA pairs across 10 languages and eight domains. Experiments show that DocSkill with Qwen3‑VL‑8B outperforms baselines on MMHop‑DocBench and performs competitively on related benchmarks.
By Jing Huang, Zhiya Tan, Changtao Miao, Shutao Gong, Yang Wang, Joey Tianyi Zhou, Jianshu Li, Ying Yan
GEPA-Curator is a framework that enhances prompt optimization for large language model agents by selecting which agent experiences should inform the optimizer. It includes task curation to filter out noisy or unreliable tasks and prior success retrieval to provide successful trajectories as context. Experiments on GoBrowse, ALFWorld, AppWorld, and τ‑bench show that GEPA-Curator matches or surpasses baseline performance, achieving up to 20.7% improvement over GEPA and 21.2% over ACE.
By Zack Yu, Nicholas Lee, Minjae Lee, Lakshya Agarwal, Luca Manolache, Wonjun Kang, Matei Zaharia, Michael W. Mahoney, Kurt Keutzer, Sehoon Kim, Amir Gholami
The paper investigates how agentic life‑science benchmarks assess reasoning over evidence by independently varying the identity of a named entity and the evidence provided. Experiments on a clinical‑outcome benchmark and tasks from Biomni and LAB‑Bench show that blinding the identity has little effect when the prompt contains decisive evidence, but dramatically reduces performance when the prompt is mainly a name and candidate list. Across 96 LAB‑Bench LitQA2 questions, the performance gap widens under a fixed passage‑redaction scheme, and a no‑evidence recall probe on the clinical benchmark matches prediction accuracy and recovers a database‑specific programme count.
By Danai Brilli, Maria Georgila
The paper introduces SceneEvolve, a framework that represents 3D scenes as executable code and improves reconstruction by evolving a harness around a small open‑source agent. SceneEvolve combines a compositional semantic action interface (SceneAct) with a harness evolution process (SceneHarness‑RSI) that uses a stronger teacher agent for iterative refinement. The resulting harness allows the small Qwen3.8‑27B model to reconstruct unseen scenes without parameter updates, achieving higher reconstruction quality and lower inference time than larger proprietary models across multiple datasets.
By Haoyu Yao, Chenguo Lin, Ruize Du, Yadong Mu
FastBEVFusion is a lightweight fusion system for real‑time 3D object detection that combines LiDAR and camera data without depth estimation. It projects modality‑specific features into semi‑aligned BEV spaces and fuses them with a dynamic cross‑attention module, achieving over 50 FPS on the nuScenes dataset. The method also demonstrates strong robustness to calibration errors.
By Tom Sanitz, Emre G\"ulsoylu, Simone Frintrop, Christian Wilms
Noise Steering is a reinforcement learning approach that jointly adjusts turbine yaw angles and power setpoints to maximize wind farm power while keeping noise within a 40 dB limit in a protected area. By learning to redirect both wakes and acoustic directivity, the method achieves only a 2–7.7 % power deficit compared to a 17 % loss with standard noise‑reduced operation. The policy remains effective when tested in higher‑fidelity simulations, showing promise for better power‑noise trade‑offs in wind farm control.
By Mart\'in de Frutos, Baptiste Corban, Claire Bizon Monroc, Ana Bu\v{s}i\'c, Donatien Dubuc, Jiamin Zhud, Esteban Ferrer
The paper introduces an autonomous AI workflow that uses specialized large language model agents to translate legacy Fortran code into performance‑portable Kokkos C++ programs. The pipeline handles translation, validation, compilation, execution, testing, debugging, and optimization, and has been shown to modernize benchmark kernels across heterogeneous GPU‑accelerated hardware. Experiments demonstrate that paid OpenAI models can produce optimized Kokkos code for only a few dollars, outperforming Fortran baselines, while open‑source models often fail to generate functional code.
By Sparsh Gupta, Kamalavasan Kamalakkannan, Maxim Moraru, Galen Shipman, Patrick Diehl